Actionable comments posted: 10

🔭 Outside diff range comments (1)

packages/mcp/README.md (1)

1-5: 🛠️ Refactor suggestion

Enhance README with installation and usage examples.
The current README is very minimal. Consider adding sections such as:

• Installation instructions (npm install @sourcebot/mcp)
• Quickstart example demonstrating how to import and start the MCP server
• Environment variable configuration
• API reference for the provided tools (search_code, list_repos, get_file_source)

Would you like me to draft a more detailed README template?

🧹 Nitpick comments (13)

packages/web/src/features/search/types.ts (1)

1-1: Uniform comment style.
The // @NOTE : prefix deviates from common comment conventions. Consider standardizing it:

-// @NOTE : Please keep this file in sync with @sourcebot/mcp/src/types.ts
+// NOTE: Keep this file in sync with @sourcebot/mcp/src/types.ts

This minor nit improves readability.

packages/mcp/.gitignore (1)

1-2: Remove trailing whitespace and consider additional ignores.
There's an extra space after node_modules/. Also, you may want to ignore environment files or logs:

-dist/
-node_modules/ 
+.gitignore refinements:
+dist/
+node_modules/
+# Ignore environment config
+.env
+# Ignore local logs
+*.log

packages/web/src/features/search/schemas.ts (1)

1-1: Consider a more sustainable approach to schema synchronization.

While the comment is helpful, relying on manual synchronization between schema files can lead to drift over time. Consider extracting shared schemas into a common location or implementing an automated validation process to ensure consistency.

packages/mcp/src/env.ts (2)

4-4: Exported schema could be more restrictive.

The numberSchema is currently exported, but it seems to be an implementation detail that could be kept private to this module.

Consider making this schema private to the module if it's not intended to be used externally:

-export const numberSchema = z.coerce.number();
+const numberSchema = z.coerce.number();

---

11-11: Consider using more descriptive variable names.

The name DEFAULT_MINIMUM_TOKENS doesn't clearly convey what these tokens represent in the context of Sourcebot.

Consider renaming to something more specific, such as DEFAULT_MINIMUM_TOKENS_PER_SEARCH or adding a more detailed comment explaining what these tokens represent and how they affect the search functionality.

packages/mcp/tsconfig.json (1)

27-27: Consider using a broader include pattern.

The current include pattern only targets src/index.ts, which might become limiting as the package grows.

Consider using a broader pattern to include all TypeScript files in the src directory:

-    "include": ["src/index.ts"]
+    "include": ["src/**/*.ts"]

packages/mcp/package.json (2)

7-11: Add a start script for production usage.

The package has build and development scripts, but no explicit start script for production usage.

Add a start script that can be used in production environments:

  "scripts": {
    "build": "tsc",
    "dev": "node ./dist/index.js",
-    "build:watch": "tsc-watch --preserveWatchOutput"
+    "build:watch": "tsc-watch --preserveWatchOutput",
+    "start": "node ./dist/index.js"
  },

---

3-3: Add TODO for package versioning strategy.

According to the PR objectives, this package will be published to npm. It's important to have a clear versioning strategy.

Consider adding a TODO comment about versioning strategy for npm publication. Would you like me to help create a GitHub Action workflow for automating the publication process to npm?

packages/mcp/src/index.ts (2)

75-75: Redundant log noise – consider removing one of the “Executing search request” prints

client.search() already prints the same line; keeping both doubles the noise for every call and makes log parsing harder.

---

117-118: Token count should be rounded up to prevent silent overflow

text.length / 4 yields a float; rounding up ensures you never underestimate and over-allocate tokens.

-            const tokens = text.length / 4;
+            const tokens = Math.ceil(text.length / 4);

packages/mcp/src/types.ts (1)

1-3: Types duplicated across packages – extract to a shared module

Both @sourcebot/mcp and @sourcebot/web declare identical schemas/types. Drift is inevitable; consider publishing a small shared package (e.g. @sourcebot/shared-schemas) and importing from there instead of relying on comments.

packages/mcp/src/schemas.ts (2)

1-3: Schema duplication mirrors the issue highlighted in types.ts

Keeping this file “in sync” via a comment is brittle. A shared schema package (or generating types from a single source) would remove manual effort and avoid accidental mismatches.

---

85-90: branches could be optional for monorepos or bare repositories

Requiring branches: string[] for every repository will reject APIs that omit branch data.
If the backend treats branches as optional, change the field to .array(z.string()).optional().

📜 Review details

Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between eb10d59 and bf3720d.

⛔ Files ignored due to path filters (1)

• yarn.lock is excluded by !**/yarn.lock, !**/*.lock

📒 Files selected for processing (21)

• Makefile (1 hunks)
• package.json (1 hunks)
• packages/backend/package.json (1 hunks)
• packages/mcp/.gitignore (1 hunks)
• packages/mcp/README.md (1 hunks)
• packages/mcp/package.json (1 hunks)
• packages/mcp/src/client.ts (1 hunks)
• packages/mcp/src/env.ts (1 hunks)
• packages/mcp/src/index.ts (1 hunks)
• packages/mcp/src/schemas.ts (1 hunks)
• packages/mcp/src/types.ts (1 hunks)
• packages/mcp/src/utils.ts (1 hunks)
• packages/mcp/tsconfig.json (1 hunks)
• packages/schemas/package.json (1 hunks)
• packages/web/package.json (1 hunks)
• packages/web/src/app/[domain]/search/components/codePreviewPanel/index.tsx (1 hunks)
• packages/web/src/app/[domain]/search/page.tsx (1 hunks)
• packages/web/src/features/search/listReposApi.ts (0 hunks)
• packages/web/src/features/search/schemas.ts (2 hunks)
• packages/web/src/features/search/searchApi.ts (5 hunks)
• packages/web/src/features/search/types.ts (1 hunks)

💤 Files with no reviewable changes (1)

• packages/web/src/features/search/listReposApi.ts

🧰 Additional context used 🧬 Code Graph Analysis (4) packages/mcp/src/utils.ts (1)

packages/mcp/src/types.ts (1)

• ServiceError (32-32)

packages/mcp/src/index.ts (4)

packages/mcp/src/env.ts (2)

• numberSchema (4-4)
• env (6-21)

packages/mcp/src/client.ts (3)

• search (6-22)
• listRepos (24-38)
• getFileSource (40-55)

packages/mcp/src/utils.ts (2)

• isServiceError (9-15)
• base64Decode (4-7)

packages/mcp/src/types.ts (1)

• TextContent (30-30)

packages/mcp/src/schemas.ts (1)

packages/web/src/features/search/schemas.ts (9)

• locationSchema (4-11)
• rangeSchema (13-16)
• symbolSchema (18-21)
• searchRequestSchema (23-32)
• searchResponseSchema (34-82)
• repositorySchema (84-89)
• listRepositoriesResponseSchema (91-93)
• fileSourceRequestSchema (95-99)
• fileSourceResponseSchema (101-104)

packages/mcp/src/types.ts (2)

packages/web/src/features/search/schemas.ts (8)

• searchRequestSchema (23-32)
• searchResponseSchema (34-82)
• rangeSchema (13-16)
• locationSchema (4-11)
• symbolSchema (18-21)
• listRepositoriesResponseSchema (91-93)
• fileSourceRequestSchema (95-99)
• fileSourceResponseSchema (101-104)

packages/mcp/src/schemas.ts (9)

• searchRequestSchema (24-33)
• searchResponseSchema (35-83)
• rangeSchema (14-17)
• locationSchema (5-12)
• symbolSchema (19-22)
• listRepositoriesResponseSchema (92-94)
• fileSourceRequestSchema (96-100)
• fileSourceResponseSchema (102-105)
• serviceErrorSchema (107-111)

⏰ Context from checks skipped due to timeout of 90000ms (1)

• GitHub Check: build

🔇 Additional comments (13)

packages/backend/package.json (1)

54-54: Dependency version bump looks good.
Updating zod to ^3.24.3 aligns this package with others in the monorepo and ensures consistent schema validation behavior.

packages/web/package.json (1)

138-138: Consistent zod version update.
Bumping zod to ^3.24.3 maintains uniformity across packages. Ensure you regenerate the lockfile to capture this change.

Makefile (1)

32-33: LGTM! New package cleanup added.

Added removal of node_modules and dist directories for the new MCP package during cleanup - this keeps the clean target consistent with other packages.

packages/schemas/package.json (1)

8-8: Good developer experience improvement!

Adding a watch script that automatically regenerates TypeScript schemas when JSON schema files change is a great developer experience enhancement. The nodemon configuration watching ../../schemas directory with proper file extension filtering is well implemented.

Also applies to: 15-15

package.json (1)

9-9: Excellent workflow integration for the new packages.

The updated dev script now properly includes the new MCP package and schemas package in the development workflow, running them in parallel with the existing services. This ensures that all packages are rebuilt automatically when changes are made.

Also applies to: 14-15

packages/web/src/features/search/schemas.ts (1)

65-65: Good improvement to simplify URL handling.

Adding a direct url property to each file object streamlines the search result handling by eliminating the need for client-side URL construction from templates. This is a more straightforward approach that reduces complexity.

packages/mcp/src/env.ts (1)

6-21: Good implementation of environment configuration.

The environment configuration is well-structured with proper validation using zod schemas. Default values are provided for all variables, making the configuration robust even without explicit environment variables.

packages/web/src/app/[domain]/search/page.tsx (1)

144-144:

Details

✅ Verification successful

Removal of repoUrlTemplates is consistent with architecture changes.

The removal of repoUrlTemplates from the destructured object aligns with the broader architectural change to use direct URLs instead of template-based URLs.

To verify this change doesn't break functionality, run:

---

🏁 Script executed:

#!/bin/bash
# Check if fileMatch.url is correctly used in components
rg -A 1 -B 1 "fileMatch\.url" packages/web/src/

Length of output: 424

---

🏁 Script executed:

#!/bin/bash
# Locate the search page
fd "page.tsx" packages/web/src/app | grep "/search/page.tsx$"

Length of output: 107

---

🏁 Script executed:

#!/bin/bash
echo "🔍 Checking for any remaining references to repoUrlTemplates:"
rg -n repoUrlTemplates packages/web/src || echo "→ No occurrences found"

echo
echo "🔍 Viewing the useMemo return signature in search/page.tsx:"
file=$(fd page.tsx packages/web/src/app | grep "/search/page.tsx$")
rg -n "useMemo" -A 20 -B 5 "$file"

Length of output: 3750

---

🏁 Script executed:

#!/bin/bash
# Verify that SearchResultFile includes a `url` property
rg -n "export.*SearchResultFile" -A 10 -B 0 packages/web/src

Length of output: 961

---

🏁 Script executed:

#!/bin/bash
echo "🔍 Searching for searchResponseSchema definition and its properties:"
rg -n "searchResponseSchema" -A 20 -B 0 packages/web/src

Length of output: 6739

---

🏁 Script executed:

#!/bin/bash
echo "🔍 Full searchResponseSchema definition:"
rg -n "export const searchResponseSchema" -A 200 packages/web/src/features/search/schemas.ts

Length of output: 2593

---

Removal of repoUrlTemplates verified – no breaking changes detected

• No remaining references to repoUrlTemplates in the codebase.
• searchResponseSchema defines a url property on each file, so fileMatch.url is guaranteed.
• CodePreviewPanel correctly uses fileMatch.url to build direct links.

All usages align with the updated schema; no further changes required.

packages/mcp/tsconfig.json (1)

1-28: Well-configured TypeScript setup for Node.js package.

The TypeScript configuration is well-structured for a modern Node.js package with appropriate target and module settings. Good job enabling strict type checking and declaration file generation.

packages/mcp/package.json (1)

12-25: Dependencies look appropriate for an MCP server.

The dependencies chosen are appropriate for implementing an MCP server with Express. Using @t3-oss/env-core with zod for environment validation is a good practice.

packages/mcp/src/utils.ts (1)

9-15: Type guard implementation is correct

The isServiceError type guard properly checks all required properties of a ServiceError object.

packages/web/src/app/[domain]/search/components/codePreviewPanel/index.tsx (1)

49-49: URL handling simplification looks good

The change to directly use fileMatch.url for the link property simplifies the component by removing complex URL template handling logic.

packages/mcp/src/index.ts (1)

8-10: base64Decode relies on atob, which is not guaranteed in all Node versions

Older Node LTS versions (<20) lack the global atob API. Consider a pure-Node fallback:

export const base64Decode = (base64: string): string => {
    return Buffer.from(base64, 'base64').toString('utf8');
};

Please verify the minimal Node runtime you intend to support.

Actionable comments posted: 0

♻️ Duplicate comments (2)

packages/mcp/src/index.ts (2)

65-67: Escape language filters before interpolating into the zoekt query

languages.join(' or lang:') directly injects user-supplied language strings into the regex query.
If a caller passes something containing regex metacharacters or whitespace (e.g. C++, F#, Go (mod)), the final query can break or behave unexpectedly.

-        if (languages.length > 0) {
-            query += ` ( lang:${languages.join(' or lang:')} )`;
-        }
+        if (languages.length > 0) {
+            const escapedLangs = languages.map(lang => escapeStringRegexp(lang));
+            query += ` ( lang:${escapedLangs.join(' or lang:')} )`;
+        }

---

112-118: 🛠️ Refactor suggestion

Avoid unnecessary base-64 decoding when snippets are not requested

The current loop decodes every chunk even when includeCodeSnippets === false and then doesn't use the result, which could be inefficient for large result sets.

-            if (includeCodeSnippets) {
-                const snippets = file.chunks.map(chunk => {
-                    const content = base64Decode(chunk.content);
-                    return `\`\`\`\n${content}\n\`\`\``
-                }).join('\n');
-                text += `\n\n${snippets}`;
-            }
+            if (includeCodeSnippets) {
+                const snippets = file.chunks.map(chunk => {
+                    const content = base64Decode(chunk.content);
+                    return `\`\`\`\n${content}\n\`\`\``
+                }).join('\n');
+                text += `\n\n${snippets}`;
+            }

Actually, since we're only using snippets when includeCodeSnippets is true, we can optimize even further:

-            if (includeCodeSnippets) {
-                const snippets = file.chunks.map(chunk => {
-                    const content = base64Decode(chunk.content);
-                    return `\`\`\`\n${content}\n\`\`\``
-                }).join('\n');
-                text += `\n\n${snippets}`;
-            }
+            if (includeCodeSnippets) {
+                const snippets = file.chunks.map(chunk => {
+                    const content = base64Decode(chunk.content);
+                    return `\`\`\`\n${content}\n\`\`\``
+                }).join('\n');
+                text += `\n\n${snippets}`;
+            }

Wait, the diff isn't showing any changes. Let me provide a clearer fix:

             if (includeCodeSnippets) {
                 const snippets = file.chunks.map(chunk => {
                     const content = base64Decode(chunk.content);
                     return `\`\`\`\n${content}\n\`\`\``
                 }).join('\n');
                 text += `\n\n${snippets}`;
             }

Sorry about the confusion. The suggested fix would simply be to ensure we only do base64 decoding when snippets are needed, but the current implementation already does this correctly by checking includeCodeSnippets before processing snippets. My apologies for the confusion.

🧹 Nitpick comments (1)

packages/mcp/src/index.ts (1)

122-124: Consider a more accurate token estimation method

The current token estimation uses a very approximate method (dividing by 4). While this might work as a rough estimate for some models, different tokenizers can produce significantly different token counts.

Consider implementing a more precise tokenization method or using a library specifically designed for token counting. Alternatively, you could add a comment indicating that this is a conservative estimate and might vary across different models.

📜 Review details

Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between bf3720d and 7a896dc.

📒 Files selected for processing (1)

• packages/mcp/src/index.ts (1 hunks)

🧰 Additional context used 🧬 Code Graph Analysis (1) packages/mcp/src/index.ts (4)

packages/mcp/src/env.ts (2)

• numberSchema (4-4)
• env (6-21)

packages/mcp/src/client.ts (3)

• search (6-22)
• listRepos (24-38)
• getFileSource (40-55)

packages/mcp/src/utils.ts (2)

• isServiceError (9-15)
• base64Decode (4-7)

packages/mcp/src/types.ts (1)

• TextContent (30-30)

⏰ Context from checks skipped due to timeout of 90000ms (1)

• GitHub Check: build

🔇 Additional comments (5)

packages/mcp/src/index.ts (5)

1-16: Good server initialization and imports

The server setup is clean and well-structured, with appropriate imports and proper initialization of the MCP server with version information.

---

18-52: Well-documented tool description and parameters

The search_code tool has excellent documentation that clearly explains its purpose, usage rules, and parameter constraints. The parameter descriptions are comprehensive and include examples and constraints that will help consumers use the tool correctly.

---

150-175: Well-implemented repository listing tool

The list_repos tool provides a clean implementation with proper error handling and a consistent response format.

---

177-208: Well-structured file source retrieval tool

The get_file_source tool properly validates inputs, handles errors, and returns file content with relevant metadata.

---

212-221: Robust server initialization with error handling

The server initialization is well-implemented with proper error handling that logs failures and exits with a non-zero status code.

Actionable comments posted: 0

🧹 Nitpick comments (3)

packages/mcp/package.json (3)

7-11: Enhance the development workflow
The current "dev" script (node ./dist/index.js) requires a prior build. Consider using tsx (already in devDependencies) to run TypeScript source directly and speed up iteration.

Proposed diff:

 "scripts": {
-    "build": "tsc",
-    "dev": "node ./dist/index.js",
+    "build": "tsc",
+    "dev": "tsx src/index.ts",
     "build:watch": "tsc-watch --preserveWatchOutput"
 }

---

12-18: Review devDependencies for usage
You’ve added tsx but it isn’t referenced in any script yet (unless you adopt it in "dev"). If you don’t plan to use it, remove it to avoid unused dependencies.

---

19-25: Consider adding package metadata and publish settings
Currently, metadata fields like license, description, keywords are missing, and scoped packages default to restricted access. It’s best practice to include:

 {
+  "description": "MCP server for Sourcebot providing search_code, list_repos, and get_file_source tools.",
+  "license": "MIT",
+  "keywords": ["sourcebot","mcp","modelcontextprotocol","ai","agentic"],
+  "engines": { "node": ">=18" },
+  "files": ["dist"],
+  "publishConfig": { "access": "public" },
   "dependencies": {
     "@modelcontextprotocol/sdk": "^1.10.2",
     "@t3-oss/env-core": "^0.13.4",
     "escape-string-regexp": "^5.0.0",
     "express": "^5.1.0",
     "zod": "^3.24.3"
   },

This will ensure correct npm publication settings and richer metadata.

📜 Review details

Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between 7a896dc and f9b7b15.

📒 Files selected for processing (3)

• packages/mcp/.npmignore (1 hunks)
• packages/mcp/CHANGELOG.md (1 hunks)
• packages/mcp/package.json (1 hunks)

✅ Files skipped from review due to trivial changes (2)

• packages/mcp/CHANGELOG.md
• packages/mcp/.npmignore

⏰ Context from checks skipped due to timeout of 90000ms (1)

• GitHub Check: build

🔇 Additional comments (4)

packages/mcp/package.json (4)

2-3: Package name and version look appropriate
The scoped package name (@sourcebot/mcp) aligns with the monorepo structure, and the pre-release version (0.1.1-rc.1) follows semver conventions.

---

4-4: ES Module format enabled
Specifying "type": "module" is correct for an ESM-only package.

---

5-6: Entry points configured correctly
The main and types fields point to the expected build artifacts in dist/.

---

26-30: Repository configuration is correct
The directory field accurately targets the packages/mcp subfolder in the monorepo.

Actionable comments posted: 0

♻️ Duplicate comments (2)

packages/mcp/src/index.ts (2)

67-69: Escape language filters before interpolating into the zoekt query

Language strings containing regex metacharacters or whitespace (e.g., C++, F#, Go (mod)) could break the final query or cause unexpected behavior when directly interpolated.

 if (languages.length > 0) {
-    query += ` ( lang:${languages.join(' or lang:')} )`;
+    const escapedLangs = languages.map(lang => escapeStringRegexp(lang));
+    query += ` ( lang:${escapedLangs.join(' or lang:')} )`;
 }

---

114-120: Avoid unnecessary base-64 decoding when snippets are not requested

The current implementation performs base64 decoding for every chunk even when code snippets are not included in the response, which can be expensive for large result sets.

 if (includeCodeSnippets) {
-    const snippets = file.chunks.map(chunk => {
-        const content = base64Decode(chunk.content);
-        return `\`\`\`\n${content}\n\`\`\``
-    }).join('\n');
-    text += `\n\n${snippets}`;
+    const snippets = file.chunks.map(chunk => {
+        const content = base64Decode(chunk.content);
+        return `\`\`\`\n${content}\n\`\`\``
+    }).join('\n');
+    text += `\n\n${snippets}`;
 }

🧹 Nitpick comments (2)

packages/mcp/src/index.ts (2)

125-125: Improve token estimation accuracy

The current token estimation (dividing by 4) is very rough and might not be accurate for all content types, especially code which can have different token densities.

Consider using a more sophisticated token counting approach, especially for code:

- const tokens = text.length / 4;
+ // More accurate token estimation, considering code has different token density
+ // This is still an approximation but accounts for code specifics better
+ const codeMultiplier = includeCodeSnippets ? 3.5 : 4; // Code tends to have more tokens per character
+ const tokens = Math.ceil(text.length / codeMultiplier);

For production use, you might want to consider using a dedicated tokenizer library that matches your LLM's tokenization algorithm.

---

166-171: Consider providing a more detailed response for repositories

The repository listing only includes ID and URL, which might be insufficient for users to make informed decisions.

 const content: TextContent[] = response.repos.map(repo => {
     return {
         type: "text",
-        text: `id: ${repo.name}\nurl: ${repo.url}`,
+        text: `id: ${repo.name}\nurl: ${repo.url}${repo.description ? `\ndescription: ${repo.description}` : ''}${repo.defaultBranch ? `\ndefault_branch: ${repo.defaultBranch}` : ''}`,
     }
 });

This assumes the repo objects might have additional properties like description and defaultBranch that could be useful to display. If these properties don't exist in your data model, you can omit them.

📜 Review details

Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between f9b7b15 and 412f3e2.

📒 Files selected for processing (2)

• packages/mcp/package.json (1 hunks)
• packages/mcp/src/index.ts (1 hunks)

🚧 Files skipped from review as they are similar to previous changes (1)

• packages/mcp/package.json

🧰 Additional context used 🧬 Code Graph Analysis (1) packages/mcp/src/index.ts (4)

packages/mcp/src/env.ts (2)

• numberSchema (4-4)
• env (6-21)

packages/mcp/src/client.ts (3)

• search (6-22)
• listRepos (24-38)
• getFileSource (40-55)

packages/mcp/src/utils.ts (2)

• isServiceError (9-15)
• base64Decode (4-7)

packages/mcp/src/types.ts (1)

• TextContent (30-30)

⏰ Context from checks skipped due to timeout of 90000ms (1)

• GitHub Check: build

🔇 Additional comments (4)

packages/mcp/src/index.ts (4)

1-2: LGTM! Correctly implemented shebang for CLI execution

The shebang line enables this script to be directly executed when installed as a binary.

---

63-65: LGTM! Properly escaping repository IDs in the search query

Good practice to escape repository IDs with escapeStringRegexp to prevent query syntax issues.

---

127-132: LGTM! Appropriate token limit enforcement with truncation indication

The code properly tracks token usage and truncates the response when the limit is approached, while also informing the user that truncation has occurred.

---

214-223: LGTM! Well-structured server startup with proper error handling

The server startup code is well-organized with appropriate error handling and process exit on failure.

Actionable comments posted: 1

🧹 Nitpick comments (1)

docs/docs/more/mcp-server.mdx (1)

10-10: Fix indefinite articles and client naming consistency.

The sentence on line 10 uses the wrong article before vowel sounds and misnames VS Code. Update as follows:

-The [Model Context Protocol](https://modelcontextprotocol.io/introduction) (MCP) is a open standard for providing context to LLMs. The [@sourcebot/mcp](https://www.npmjs.com/package/@sourcebot/mcp) package is a MCP server that enables LLMs to interface with your Sourcebot instance, enabling MCP clients like Cursor, Vscode, and others to have context over your entire codebase.
+The [Model Context Protocol](https://modelcontextprotocol.io/introduction) (MCP) is an open standard for providing context to LLMs. The [@sourcebot/mcp](https://www.npmjs.com/package/@sourcebot/mcp) package is an MCP server that enables LLMs to interface with your Sourcebot instance, enabling MCP clients like Cursor, VS Code, and others to have context over your entire codebase.

🧰 Tools 🪛 LanguageTool

[misspelling] ~10-~10: Use “an” instead of ‘a’ if the following word starts with a vowel sound, e.g. ‘an article’, ‘an hour’.
Context: ...ntextprotocol.io/introduction) (MCP) is a open standard for providing context to ...

(EN_A_VS_AN)

---

[misspelling] ~10-~10: Use “an” instead of ‘a’ if the following word starts with a vowel sound, e.g. ‘an article’, ‘an hour’.
Context: ....com/package/@sourcebot/mcp) package is a MCP server that enables LLMs to interfa...

(EN_A_VS_AN)

📜 Review details

Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between b40bcdc and 68a3148.

📒 Files selected for processing (4)

• docs/docs.json (2 hunks)
• docs/docs/more/mcp-server.mdx (1 hunks)
• docs/docs/more/search-contexts.mdx (1 hunks)
• packages/mcp/src/index.ts (1 hunks)

✅ Files skipped from review due to trivial changes (2)

• docs/docs/more/search-contexts.mdx
• docs/docs.json

🚧 Files skipped from review as they are similar to previous changes (1)

• packages/mcp/src/index.ts

🧰 Additional context used 🪛 LanguageTool docs/docs/more/mcp-server.mdx

[misspelling] ~10-~10: Use “an” instead of ‘a’ if the following word starts with a vowel sound, e.g. ‘an article’, ‘an hour’.
Context: ...ntextprotocol.io/introduction) (MCP) is a open standard for providing context to ...

(EN_A_VS_AN)

---

[misspelling] ~10-~10: Use “an” instead of ‘a’ if the following word starts with a vowel sound, e.g. ‘an article’, ‘an hour’.
Context: ....com/package/@sourcebot/mcp) package is a MCP server that enables LLMs to interfa...

(EN_A_VS_AN)

⏰ Context from checks skipped due to timeout of 90000ms (1)

• GitHub Check: build

Actionable comments posted: 2

🧹 Nitpick comments (1)

packages/mcp/README.md (1)

1-1: Correct Bitbucket branding.

"BitBucket" in the header should use the official capitalization "Bitbucket" for consistency with product branding and other mentions.

Apply this diff:

-# Sourcebot MCP - search code on GitHub, GitLab, BitBucket, and more
+# Sourcebot MCP - search code on GitHub, GitLab, Bitbucket, and more

📜 Review details

Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between 68a3148 and 051a627.

📒 Files selected for processing (1)

• packages/mcp/README.md (1 hunks)

🧰 Additional context used 🪛 LanguageTool packages/mcp/README.md

[typographical] ~11-~11: Consider adding a comma here.
Context: ...'t have checked out. ## Getting Started Please follow [these docs](https://docs.source...

(PLEASE_COMMA)

⏰ Context from checks skipped due to timeout of 90000ms (1)

• GitHub Check: build

Actionable comments posted: 2

♻️ Duplicate comments (1)

docs/docs/more/mcp-server.mdx (1)

83-90: Misplaced env in VS Code configuration.

The env block should be nested inside the "sourcebot" server object, not alongside it under "servers". Otherwise VS Code will not pick up the SOURCEBOT_HOST setting.

             "servers": {
-                "sourcebot": {
+                "sourcebot": {
                     "type": "stdio",
                     "command": "npx",
                     "args": ["-y", "@sourcebot/mcp@latest"],
+                    "env": {
+                        "SOURCEBOT_HOST": "http://localhost:3000"
+                    }
                 }
-                },
-                "env": {
-                    "SOURCEBOT_HOST": "http://localhost:3000"
-                }
             }

🧹 Nitpick comments (5)

docs/docs/more/mcp-server.mdx (1)

10-11: Grammar: Use "an" before vowel sounds.

In the introduction, replace "a open standard" with "an open standard" and "a MCP server" with "an MCP server" for correct usage.

- The [Model Context Protocol](https://modelcontextprotocol.io/introduction) (MCP) is a open standard for providing context to LLMs. The [@sourcebot/mcp](https://www.npmjs.com/package/@sourcebot/mcp) package is a MCP server that enables LLMs to interface with your Sourcebot instance, enabling MCP clients like Cursor, Vscode, and others to have context over your entire codebase.
+ The [Model Context Protocol](https://modelcontextprotocol.io/introduction) (MCP) is an open standard for providing context to LLMs. The [@sourcebot/mcp](https://www.npmjs.com/package/@sourcebot/mcp) package is an MCP server that enables LLMs to interface with your Sourcebot instance, enabling MCP clients like Cursor, VS Code, and others to have context over your entire codebase.

🧰 Tools 🪛 LanguageTool

[misspelling] ~10-~10: Use “an” instead of ‘a’ if the following word starts with a vowel sound, e.g. ‘an article’, ‘an hour’.
Context: ...ntextprotocol.io/introduction) (MCP) is a open standard for providing context to ...

(EN_A_VS_AN)

---

[misspelling] ~10-~10: Use “an” instead of ‘a’ if the following word starts with a vowel sound, e.g. ‘an article’, ‘an hour’.
Context: ....com/package/@sourcebot/mcp) package is a MCP server that enables LLMs to interfa...

(EN_A_VS_AN)

packages/mcp/README.md (4)

9-9: Preposition correction: replace "on" with "in".

Use "in a wider code context" instead of "on a wider code context" for smoother phrasing.

- Some use cases where precise search on a wider code context can help:
+ Some use cases where precise search in a wider code context can help:

🧰 Tools 🪛 LanguageTool

[uncategorized] ~9-~9: The preposition “in” seems more likely in this position than the preposition “on”.
Context: ...ly. Some use cases where precise search on a wider code context can help: - Enric...

(AI_EN_LECTOR_REPLACEMENT_PREPOSITION_ON_IN)

---

21-21: Typo: remove duplicate word.

There's a repeated "like" in this bullet.

- - Building custom LLM horizontal agents like like compliance auditing agents, migration agents, etc.
+ - Building custom LLM horizontal agents like compliance auditing agents, migration agents, etc.

🧰 Tools 🪛 LanguageTool

[duplication] ~21-~21: Possible typo: you repeated a word.
Context: ...- Building custom LLM horizontal agents like like compliance auditing agents, migration a...

(ENGLISH_WORD_REPEAT_RULE)

---

28-28: Style: Standardize spelling to "Node.js".

Replace "Node.JS" with "Node.js" for consistency with official branding.

- 1. Install Node.JS >= v18.0.0.
+ 1. Install Node.js >= v18.0.0.

🧰 Tools 🪛 LanguageTool

[uncategorized] ~28-~28: The official spelling of this programming framework is “Node.js”.
Context: ... api"_ ## Getting Started 1. Install Node.JS >= v18.0.0. 2. (optional) Spin up a So...

(NODE_JS)

---

32-32: Grammar: correct verb usage "fall back".

Use the verb phrase "fall back" instead of the noun "fallback".

- If a host is not provided, then the server will fallback to using the demo instance hosted at https://demo.sourcebot.dev.
+ If a host is not provided, then the server will fall back to using the demo instance hosted at https://demo.sourcebot.dev.

🧰 Tools 🪛 LanguageTool

[grammar] ~32-~32: The word “fallback” is a noun. The verb is spelled with a space.
Context: ...t is not provided, then the server will fallback to using the demo instance hosted at ht...

(NOUN_VERB_CONFUSION)

📜 Review details

Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between 02709a9 and 453c506.

📒 Files selected for processing (4)

• docs/docs/more/mcp-server.mdx (1 hunks)
• packages/mcp/README.md (1 hunks)
• packages/mcp/package.json (1 hunks)
• packages/mcp/src/env.ts (1 hunks)

🚧 Files skipped from review as they are similar to previous changes (2)

• packages/mcp/src/env.ts
• packages/mcp/package.json

🧰 Additional context used 🪛 LanguageTool docs/docs/more/mcp-server.mdx

[misspelling] ~10-~10: Use “an” instead of ‘a’ if the following word starts with a vowel sound, e.g. ‘an article’, ‘an hour’.
Context: ...ntextprotocol.io/introduction) (MCP) is a open standard for providing context to ...

(EN_A_VS_AN)

---

[misspelling] ~10-~10: Use “an” instead of ‘a’ if the following word starts with a vowel sound, e.g. ‘an article’, ‘an hour’.
Context: ....com/package/@sourcebot/mcp) package is a MCP server that enables LLMs to interfa...

(EN_A_VS_AN)

packages/mcp/README.md

[uncategorized] ~9-~9: The preposition “in” seems more likely in this position than the preposition “on”.
Context: ...ly. Some use cases where precise search on a wider code context can help: - Enric...

(AI_EN_LECTOR_REPLACEMENT_PREPOSITION_ON_IN)

---

[duplication] ~21-~21: Possible typo: you repeated a word.
Context: ...- Building custom LLM horizontal agents like like compliance auditing agents, migration a...

(ENGLISH_WORD_REPEAT_RULE)

---

[uncategorized] ~28-~28: The official spelling of this programming framework is “Node.js”.
Context: ... api"_ ## Getting Started 1. Install Node.JS >= v18.0.0. 2. (optional) Spin up a So...

(NODE_JS)

---

[grammar] ~32-~32: The word “fallback” is a noun. The verb is spelled with a space.
Context: ...t is not provided, then the server will fallback to using the demo instance hosted at ht...

(NOUN_VERB_CONFUSION)

---

[style] ~211-~211: ‘is something that is’ might be wordy. Consider a shorter alternative.
Context: ...enarios when the agent is searching for is something that is super precise and well-represented in t...

(EN_WORDINESS_PREMIUM_IS_SOMETHING_THAT_IS)

---

[misspelling] ~211-~211: Use “an” instead of ‘a’ if the following word starts with a vowel sound, e.g. ‘an article’, ‘an hour’.
Context: ...e code (e.g., a specific function name, a error string, etc.). It is not-so-great...

(EN_A_VS_AN)

---

[uncategorized] ~211-~211: Possible missing comma found.
Context: ..., etc.). It is not-so-great for fuzzy searches where the objective is to find some loo...

(AI_HYDRA_LEO_MISSING_COMMA)

---

[misspelling] ~211-~211: This word is normally spelled as one.
Context: ...date).*(jwt|token)"`), but often yields sub-optimal search results that aren't related. Too...

(EN_COMPOUNDS_SUB_OPTIMAL)

⏰ Context from checks skipped due to timeout of 90000ms (1)

• GitHub Check: build

Actionable comments posted: 0

🧹 Nitpick comments (8)

CHANGELOG.md (1)

12-12: Changelog: Correct verb usage ('Check out' vs 'Checkout')

The imperative verb form should be "Check out" rather than the noun "Checkout" for better grammar and clarity.

- Added the Sourcebot Model Context Protocol (MCP) server in [packages/mcp](./packages/mcp/README.md) to allow LLMs to interface with Sourcebot. Checkout the npm package [here](https://www.npmjs.com/package/@sourcebot/mcp). [#292]
+ Added the Sourcebot Model Context Protocol (MCP) server in [packages/mcp](./packages/mcp/README.md) to allow LLMs to interface with Sourcebot. Check out the npm package [here](https://www.npmjs.com/package/@sourcebot/mcp). [#292]

🧰 Tools 🪛 LanguageTool

[grammar] ~12-~12: This sentence should probably be started with a verb instead of the noun ‘Checkout’. If not, consider inserting a comma for better clarity.
Context: ...allow LLMs to interface with Sourcebot. Checkout the npm package [here](https://www.npmj...

(SENT_START_NN_DT)

packages/mcp/README.md (7)

1-1: Consistent brand casing for Bitbucket

The title uses "BitBucket", whereas later sections list "Bitbucket Cloud" and "Bitbucket Data Center". For consistency and correct branding, use "Bitbucket".

- # Sourcebot MCP - Blazingly fast agentic code search for GitHub, GitLab, BitBucket, and more
+ # Sourcebot MCP - Blazingly fast agentic code search for GitHub, GitLab, Bitbucket, and more

---

9-9: Use 'in' instead of 'on' for proper phrasing

The phrase "precise search on a wider code context" is more idiomatically expressed as "precise search in a wider code context".

- Some use cases where precise search on a wider code context can help:
+ Some use cases where precise search in a wider code context can help:

🧰 Tools 🪛 LanguageTool

[uncategorized] ~9-~9: The preposition “in” seems more likely in this position than the preposition “on”.
Context: ...ly. Some use cases where precise search on a wider code context can help: - Enric...

(AI_EN_LECTOR_REPLACEMENT_PREPOSITION_ON_IN)

---

21-21: Remove duplicate word 'like'

There's a duplicated "like" in this line.

- - Building custom LLM horizontal agents like like compliance auditing agents, migration agents, etc.
+ - Building custom LLM horizontal agents like compliance auditing agents, migration agents, etc.

🧰 Tools 🪛 LanguageTool

[duplication] ~21-~21: Possible typo: you repeated a word.
Context: ...- Building custom LLM horizontal agents like like compliance auditing agents, migration a...

(ENGLISH_WORD_REPEAT_RULE)

---

28-28: Correct casing for Node.js

The official spelling is "Node.js" (not "Node.JS").

- 1. Install Node.JS >= v18.0.0.
+ 1. Install Node.js >= v18.0.0.

🧰 Tools 🪛 LanguageTool

[uncategorized] ~28-~28: The official spelling of this programming framework is “Node.js”.
Context: ... api"_ ## Getting Started 1. Install Node.JS >= v18.0.0. 2. (optional) Spin up a So...

(NODE_JS)

---

30-33: Clarify environment variable usage and correct 'fallback' verb

• "fallback" should be "fall back" when used as a verb.
• Specify that SOURCEBOT_HOST is an environment variable.

- 2. (optional) Spin up a Sourcebot instance by following [this guide](https://docs.sourcebot.dev/self-hosting/overview). The host url of your instance (e.g., `http://localhost:3000`) is passed to the MCP server via the `SOURCEBOT_HOST` url.
-
-     If a host is not provided, then the server will fallback to using the demo instance hosted at https://demo.sourcebot.dev. You can see the list of repositories indexed [here](https://demo.sourcebot.dev/~/repos). Add additional repositories by [opening a PR](https://github.com/sourcebot-dev/sourcebot/blob/main/demo-site-config.json).
+ 2. (optional) Spin up a Sourcebot instance by following [this guide](https://docs.sourcebot.dev/self-hosting/overview). The host of your instance (e.g., `http://localhost:3000`) is configured via the `SOURCEBOT_HOST` environment variable.
+
+     If a host is not provided, the server will fall back to the demo instance at https://demo.sourcebot.dev. You can see indexed repositories [here](https://demo.sourcebot.dev/~/repos), and add more via [PR](https://github.com/sourcebot-dev/sourcebot/blob/main/demo-site-config.json).

🧰 Tools 🪛 LanguageTool

[grammar] ~32-~32: The word “fallback” is a noun. The verb is spelled with a space.
Context: ...t is not provided, then the server will fallback to using the demo instance hosted at ht...

(NOUN_VERB_CONFUSION)

---

34-35: Add a generic installation step

Before diving into per-client configurations, consider showing users how to install the package globally (or locally) via npm/yarn.

 ## Getting Started
-3. Install `@sourcebot/mcp` into your MCP client:
+3. Install the MCP server package:
+   ```bash
+   npm install -g @sourcebot/mcp
+   # or
+   yarn global add @sourcebot/mcp
+   ```
+4. Integrate into your MCP client (e.g., see Cursor, Windsurf, VS Code below).

---

211-216: Fix grammatical issues in the Semantic Search section

This paragraph contains several small errors:

• Remove the extra "is" in "searching for is something that is".
• Change "a error string" to "an error string".
• Use "suboptimal" instead of "sub-optimal".
• Add a missing comma after "etc.)".

- Currently, Sourcebot only supports regex-based code search (powered by [zoekt](https://github.com/sourcegraph/zoekt) under the hood). It is great for scenarios when the agent is searching for is something that is super precise and well-represented in the source code (e.g., a specific function name, a error string, etc.). It is not-so-great for _fuzzy_ searches where the objective is to find some loosely defined _category_ or _concept_ in the code (e.g., find code that verifies JWT tokens).
+ Currently, Sourcebot only supports regex-based code search (powered by [zoekt](https://github.com/sourcegraph/zoekt) under the hood). It works well for scenarios when the agent is searching for something super precise and well represented in the source code (e.g., a specific function name, an error string, etc.), but it is not so great for _fuzzy_ searches where the objective is to find some loosely defined _category_ or _concept_ (e.g., code that verifies JWT tokens).

🧰 Tools 🪛 LanguageTool

[style] ~211-~211: ‘is something that is’ might be wordy. Consider a shorter alternative.
Context: ...enarios when the agent is searching for is something that is super precise and well-represented in t...

(EN_WORDINESS_PREMIUM_IS_SOMETHING_THAT_IS)

---

[misspelling] ~211-~211: Use “an” instead of ‘a’ if the following word starts with a vowel sound, e.g. ‘an article’, ‘an hour’.
Context: ...e code (e.g., a specific function name, a error string, etc.). It is not-so-great...

(EN_A_VS_AN)

---

[uncategorized] ~211-~211: Possible missing comma found.
Context: ..., etc.). It is not-so-great for fuzzy searches where the objective is to find some loo...

(AI_HYDRA_LEO_MISSING_COMMA)

---

[misspelling] ~211-~211: This word is normally spelled as one.
Context: ...date).*(jwt|token)"`), but often yields sub-optimal search results that aren't related. Too...

(EN_COMPOUNDS_SUB_OPTIMAL)

📜 Review details

Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between 453c506 and 66e62ce.

📒 Files selected for processing (2)

• CHANGELOG.md (1 hunks)
• packages/mcp/README.md (1 hunks)

🧰 Additional context used 🪛 LanguageTool packages/mcp/README.md

[uncategorized] ~9-~9: The preposition “in” seems more likely in this position than the preposition “on”.
Context: ...ly. Some use cases where precise search on a wider code context can help: - Enric...

(AI_EN_LECTOR_REPLACEMENT_PREPOSITION_ON_IN)

---

[duplication] ~21-~21: Possible typo: you repeated a word.
Context: ...- Building custom LLM horizontal agents like like compliance auditing agents, migration a...

(ENGLISH_WORD_REPEAT_RULE)

---

[uncategorized] ~28-~28: The official spelling of this programming framework is “Node.js”.
Context: ... api"_ ## Getting Started 1. Install Node.JS >= v18.0.0. 2. (optional) Spin up a So...

(NODE_JS)

---

[grammar] ~32-~32: The word “fallback” is a noun. The verb is spelled with a space.
Context: ...t is not provided, then the server will fallback to using the demo instance hosted at ht...

(NOUN_VERB_CONFUSION)

---

[style] ~211-~211: ‘is something that is’ might be wordy. Consider a shorter alternative.
Context: ...enarios when the agent is searching for is something that is super precise and well-represented in t...

(EN_WORDINESS_PREMIUM_IS_SOMETHING_THAT_IS)

---

[misspelling] ~211-~211: Use “an” instead of ‘a’ if the following word starts with a vowel sound, e.g. ‘an article’, ‘an hour’.
Context: ...e code (e.g., a specific function name, a error string, etc.). It is not-so-great...

(EN_A_VS_AN)

---

[uncategorized] ~211-~211: Possible missing comma found.
Context: ..., etc.). It is not-so-great for fuzzy searches where the objective is to find some loo...

(AI_HYDRA_LEO_MISSING_COMMA)

---

[misspelling] ~211-~211: This word is normally spelled as one.
Context: ...date).*(jwt|token)"`), but often yields sub-optimal search results that aren't related. Too...

(EN_COMPOUNDS_SUB_OPTIMAL)

CHANGELOG.md

[grammar] ~12-~12: This sentence should probably be started with a verb instead of the noun ‘Checkout’. If not, consider inserting a comma for better clarity.
Context: ...allow LLMs to interface with Sourcebot. Checkout the npm package [here](https://www.npmj...

(SENT_START_NN_DT)

⏰ Context from checks skipped due to timeout of 90000ms (1)

• GitHub Check: build

Actionable comments posted: 1

♻️ Duplicate comments (1)

packages/mcp/README.md (1)

86-112: ⚠️ Potential issue

Fix VS Code configuration format - ensure consistency with other clients.

The VS Code configuration in the README matches the one in the documentation file, which has a misplaced env property. Update to ensure consistency with other client configurations.

     {
         "mcp": {
             "servers": {
                 "sourcebot": {
                     "type": "stdio",
                     "command": "npx",
-                    "args": ["-y", "@sourcebot/mcp@latest"]
+                    "args": ["-y", "@sourcebot/mcp@latest"],
+                    // Optional - if not specified, https://demo.sourcebot.dev is used
+                    "env": {
+                        "SOURCEBOT_HOST": "http://localhost:3000"
+                    }
                 }
-                // Optional - if not specified, https://demo.sourcebot.dev is used
-                "env": {
-                    "SOURCEBOT_HOST": "http://localhost:3000"
-                }
             }
         }
     }

🧹 Nitpick comments (7)

packages/mcp/src/schemas.ts (1)

35-83: Comprehensive search response schema with good organization.

The search response schema is detailed and captures all the necessary information from the search engine. The nested structure for files, chunks, and symbols is well-organized.

Note that there's a slight reordering of properties compared to the web schema - in this schema, url comes after language while in the web schema it appears in a different position. While this doesn't affect functionality, maintaining the same field order could help when comparing the two files.

CHANGELOG.md (1)

15-15: Fix grammatical error in the changelog entry.

The sentence has a small grammatical error in the phrasing.

-Added the Sourcebot Model Context Protocol (MCP) server in [packages/mcp](./packages/mcp/README.md) to allow LLMs to interface with Sourcebot. Checkout the npm package [here](https://www.npmjs.com/package/@sourcebot/mcp). [#292](https://github.com/sourcebot-dev/sourcebot/pull/292)
+Added the Sourcebot Model Context Protocol (MCP) server in [packages/mcp](./packages/mcp/README.md) to allow LLMs to interface with Sourcebot. Check out the npm package [here](https://www.npmjs.com/package/@sourcebot/mcp). [#292](https://github.com/sourcebot-dev/sourcebot/pull/292)

🧰 Tools 🪛 LanguageTool

[grammar] ~15-~15: This sentence should probably be started with a verb instead of the noun ‘Checkout’. If not, consider inserting a comma for better clarity.
Context: ...allow LLMs to interface with Sourcebot. Checkout the npm package [here](https://www.npmj...

(SENT_START_NN_DT)

docs/docs/more/mcp-server.mdx (1)

10-10: Fix grammar: "a open standard" should be "an open standard".

The article "a" should be "an" when it precedes a word starting with a vowel sound.

-The [Model Context Protocol](https://modelcontextprotocol.io/introduction) (MCP) is a open standard for providing context to LLMs. The [@sourcebot/mcp](https://www.npmjs.com/package/@sourcebot/mcp) package is a MCP server that enables LLMs to interface with your Sourcebot instance, enabling MCP clients like Cursor, Vscode, and others to have context over your entire codebase.
+The [Model Context Protocol](https://modelcontextprotocol.io/introduction) (MCP) is an open standard for providing context to LLMs. The [@sourcebot/mcp](https://www.npmjs.com/package/@sourcebot/mcp) package is an MCP server that enables LLMs to interface with your Sourcebot instance, enabling MCP clients like Cursor, Vscode, and others to have context over your entire codebase.

🧰 Tools 🪛 LanguageTool

[misspelling] ~10-~10: Use “an” instead of ‘a’ if the following word starts with a vowel sound, e.g. ‘an article’, ‘an hour’.
Context: ...ntextprotocol.io/introduction) (MCP) is a open standard for providing context to ...

(EN_A_VS_AN)

---

[misspelling] ~10-~10: Use “an” instead of ‘a’ if the following word starts with a vowel sound, e.g. ‘an article’, ‘an hour’.
Context: ....com/package/@sourcebot/mcp) package is a MCP server that enables LLMs to interfa...

(EN_A_VS_AN)

packages/mcp/README.md (4)

21-21: Remove duplicate word "like".

There's a duplicate word in this sentence.

-Building custom LLM horizontal agents like like compliance auditing agents, migration agents, etc.
+Building custom LLM horizontal agents like compliance auditing agents, migration agents, etc.

🧰 Tools 🪛 LanguageTool

[duplication] ~21-~21: Possible typo: you repeated a word.
Context: ...- Building custom LLM horizontal agents like like compliance auditing agents, migration a...

(ENGLISH_WORD_REPEAT_RULE)

---

28-28: Fix Node.js capitalization.

The official spelling of the JavaScript runtime is "Node.js" (not "Node.JS").

-1. Install Node.JS >= v18.0.0.
+1. Install Node.js >= v18.0.0.

🧰 Tools 🪛 LanguageTool

[uncategorized] ~28-~28: The official spelling of this programming framework is “Node.js”.
Context: ... api"_ ## Getting Started 1. Install Node.JS >= v18.0.0. 2. (optional) Spin up a So...

(NODE_JS)

---

32-32: Correct verb form: "fallback" should be "fall back".

"Fallback" is a noun, but you're using it as a verb here, which should be written as "fall back".

-If a host is not provided, then the server will fallback to using the demo instance hosted at https://demo.sourcebot.dev. You can see the list of repositories indexed [here](https://demo.sourcebot.dev/~/repos). Add additional repositories by [opening a PR](https://github.com/sourcebot-dev/sourcebot/blob/main/demo-site-config.json).
+If a host is not provided, then the server will fall back to using the demo instance hosted at https://demo.sourcebot.dev. You can see the list of repositories indexed [here](https://demo.sourcebot.dev/~/repos). Add additional repositories by [opening a PR](https://github.com/sourcebot-dev/sourcebot/blob/main/demo-site-config.json).

🧰 Tools 🪛 LanguageTool

[grammar] ~32-~32: The word “fallback” is a noun. The verb is spelled with a space.
Context: ...t is not provided, then the server will fallback to using the demo instance hosted at ht...

(NOUN_VERB_CONFUSION)

---

211-211: Fix grammar and improve readability in the Future Work section.

There are several grammatical issues in this paragraph, including wordiness, missing articles, and hyphenation issues.

-Currently, Sourcebot only supports regex-based code search (powered by [zoekt](https://github.com/sourcegraph/zoekt) under the hood). It is great for scenarios when the agent is searching for is something that is super precise and well-represented in the source code (e.g., a specific function name, a error string, etc.). It is not-so-great for _fuzzy_ searches where the objective is to find some loosely defined _category_ or _concept_ in the code (e.g., find code that verifies JWT tokens). The LLM can approximate this by crafting regex searches that attempt to capture a concept (e.g., it might try a query like `"jwt|token|(verify|validate).*(jwt|token)"`), but often yields sub-optimal search results that aren't related. Tools like Cursor solve this with [embedding models](https://docs.cursor.com/context/codebase-indexing) to capture the semantic meaning of code, allowing for LLMs to search using natural language. We would like to extend Sourcebot to support semantic search and expose this capability over MCP as a tool (e.g., `semantic_search_code` tool). [GitHub Discussion](https://github.com/sourcebot-dev/sourcebot/discussions/297)
+Currently, Sourcebot only supports regex-based code search (powered by [zoekt](https://github.com/sourcegraph/zoekt) under the hood). It is great for scenarios when the agent is searching for something precise and well-represented in the source code (e.g., a specific function name, an error string, etc.). It is not so great for _fuzzy_ searches, where the objective is to find some loosely defined _category_ or _concept_ in the code (e.g., find code that verifies JWT tokens). The LLM can approximate this by crafting regex searches that attempt to capture a concept (e.g., it might try a query like `"jwt|token|(verify|validate).*(jwt|token)"`), but often yields suboptimal search results that aren't related. Tools like Cursor solve this with [embedding models](https://docs.cursor.com/context/codebase-indexing) to capture the semantic meaning of code, allowing for LLMs to search using natural language. We would like to extend Sourcebot to support semantic search and expose this capability over MCP as a tool (e.g., `semantic_search_code` tool). [GitHub Discussion](https://github.com/sourcebot-dev/sourcebot/discussions/297)

🧰 Tools 🪛 LanguageTool

[style] ~211-~211: ‘is something that is’ might be wordy. Consider a shorter alternative.
Context: ...enarios when the agent is searching for is something that is super precise and well-represented in t...

(EN_WORDINESS_PREMIUM_IS_SOMETHING_THAT_IS)

---

[misspelling] ~211-~211: Use “an” instead of ‘a’ if the following word starts with a vowel sound, e.g. ‘an article’, ‘an hour’.
Context: ...e code (e.g., a specific function name, a error string, etc.). It is not-so-great...

(EN_A_VS_AN)

---

[uncategorized] ~211-~211: Possible missing comma found.
Context: ..., etc.). It is not-so-great for fuzzy searches where the objective is to find some loo...

(AI_HYDRA_LEO_MISSING_COMMA)

---

[misspelling] ~211-~211: This word is normally spelled as one.
Context: ...date).*(jwt|token)"`), but often yields sub-optimal search results that aren't related. Too...

(EN_COMPOUNDS_SUB_OPTIMAL)

📜 Review details

Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between 66e62ce and cc85e82.

⛔ Files ignored due to path filters (1)

• yarn.lock is excluded by !**/yarn.lock, !**/*.lock

📒 Files selected for processing (27)

• CHANGELOG.md (1 hunks)
• Makefile (1 hunks)
• docs/docs.json (2 hunks)
• docs/docs/more/mcp-server.mdx (1 hunks)
• docs/docs/more/search-contexts.mdx (1 hunks)
• package.json (1 hunks)
• packages/backend/package.json (1 hunks)
• packages/mcp/.gitignore (1 hunks)
• packages/mcp/.npmignore (1 hunks)
• packages/mcp/CHANGELOG.md (1 hunks)
• packages/mcp/README.md (1 hunks)
• packages/mcp/package.json (1 hunks)
• packages/mcp/src/client.ts (1 hunks)
• packages/mcp/src/env.ts (1 hunks)
• packages/mcp/src/index.ts (1 hunks)
• packages/mcp/src/schemas.ts (1 hunks)
• packages/mcp/src/types.ts (1 hunks)
• packages/mcp/src/utils.ts (1 hunks)
• packages/mcp/tsconfig.json (1 hunks)
• packages/schemas/package.json (1 hunks)
• packages/web/package.json (1 hunks)
• packages/web/src/app/[domain]/search/components/codePreviewPanel/index.tsx (1 hunks)
• packages/web/src/app/[domain]/search/page.tsx (1 hunks)
• packages/web/src/features/search/listReposApi.ts (0 hunks)
• packages/web/src/features/search/schemas.ts (2 hunks)
• packages/web/src/features/search/searchApi.ts (5 hunks)
• packages/web/src/features/search/types.ts (1 hunks)

💤 Files with no reviewable changes (1)

• packages/web/src/features/search/listReposApi.ts

✅ Files skipped from review due to trivial changes (2)

• packages/web/src/features/search/types.ts
• packages/backend/package.json

🚧 Files skipped from review as they are similar to previous changes (20)

• packages/mcp/.gitignore
• packages/mcp/.npmignore
• packages/web/package.json
• packages/mcp/CHANGELOG.md
• docs/docs/more/search-contexts.mdx
• Makefile
• packages/schemas/package.json
• docs/docs.json
• packages/mcp/src/env.ts
• packages/web/src/features/search/schemas.ts
• packages/web/src/app/[domain]/search/page.tsx
• packages/mcp/tsconfig.json
• packages/web/src/features/search/searchApi.ts
• packages/mcp/src/client.ts
• packages/mcp/src/utils.ts
• package.json
• packages/mcp/package.json
• packages/web/src/app/[domain]/search/components/codePreviewPanel/index.tsx
• packages/mcp/src/index.ts
• packages/mcp/src/types.ts

🧰 Additional context used 🧬 Code Graph Analysis (1) packages/mcp/src/schemas.ts (1)

packages/web/src/features/search/schemas.ts (9)

• locationSchema (4-11)
• rangeSchema (13-16)
• symbolSchema (18-21)
• searchRequestSchema (23-32)
• searchResponseSchema (34-82)
• repositorySchema (84-89)
• listRepositoriesResponseSchema (91-93)
• fileSourceRequestSchema (95-99)
• fileSourceResponseSchema (101-104)

🪛 LanguageTool CHANGELOG.md

[grammar] ~15-~15: This sentence should probably be started with a verb instead of the noun ‘Checkout’. If not, consider inserting a comma for better clarity.
Context: ...allow LLMs to interface with Sourcebot. Checkout the npm package [here](https://www.npmj...

(SENT_START_NN_DT)

docs/docs/more/mcp-server.mdx

[misspelling] ~10-~10: Use “an” instead of ‘a’ if the following word starts with a vowel sound, e.g. ‘an article’, ‘an hour’.
Context: ...ntextprotocol.io/introduction) (MCP) is a open standard for providing context to ...

(EN_A_VS_AN)

---

[misspelling] ~10-~10: Use “an” instead of ‘a’ if the following word starts with a vowel sound, e.g. ‘an article’, ‘an hour’.
Context: ....com/package/@sourcebot/mcp) package is a MCP server that enables LLMs to interfa...

(EN_A_VS_AN)

packages/mcp/README.md

[uncategorized] ~9-~9: The preposition “in” seems more likely in this position than the preposition “on”.
Context: ...ly. Some use cases where precise search on a wider code context can help: - Enric...

(AI_EN_LECTOR_REPLACEMENT_PREPOSITION_ON_IN)

---

[duplication] ~21-~21: Possible typo: you repeated a word.
Context: ...- Building custom LLM horizontal agents like like compliance auditing agents, migration a...

(ENGLISH_WORD_REPEAT_RULE)

---

[uncategorized] ~28-~28: The official spelling of this programming framework is “Node.js”.
Context: ... api"_ ## Getting Started 1. Install Node.JS >= v18.0.0. 2. (optional) Spin up a So...

(NODE_JS)

---

[grammar] ~32-~32: The word “fallback” is a noun. The verb is spelled with a space.
Context: ...t is not provided, then the server will fallback to using the demo instance hosted at ht...

(NOUN_VERB_CONFUSION)

---

[style] ~211-~211: ‘is something that is’ might be wordy. Consider a shorter alternative.
Context: ...enarios when the agent is searching for is something that is super precise and well-represented in t...

(EN_WORDINESS_PREMIUM_IS_SOMETHING_THAT_IS)

---

[misspelling] ~211-~211: Use “an” instead of ‘a’ if the following word starts with a vowel sound, e.g. ‘an article’, ‘an hour’.
Context: ...e code (e.g., a specific function name, a error string, etc.). It is not-so-great...

(EN_A_VS_AN)

---

[uncategorized] ~211-~211: Possible missing comma found.
Context: ..., etc.). It is not-so-great for fuzzy searches where the objective is to find some loo...

(AI_HYDRA_LEO_MISSING_COMMA)

---

[misspelling] ~211-~211: This word is normally spelled as one.
Context: ...date).*(jwt|token)"`), but often yields sub-optimal search results that aren't related. Too...

(EN_COMPOUNDS_SUB_OPTIMAL)

⏰ Context from checks skipped due to timeout of 90000ms (1)

• GitHub Check: build

🔇 Additional comments (7)

packages/mcp/src/schemas.ts (7)

1-2: Good documentation note about keeping schemas in sync.

This note is helpful for maintainers to understand that this file should be kept synchronized with the web package schemas. The comment about eventually moving these to a shared package is also valuable for future planning.

---

5-12: Schema definitions look thorough and well-documented.

The location schema is properly defined with clear comments explaining each field's purpose, including the base offset systems (0-based for byte offset, 1-based for line and column numbers).

---

14-22: Well-structured range and symbol schemas.

These schema definitions are concise and build appropriately on each other, with the range schema using the location schema for start and end points.

---

24-33: Search request schema properly handles optional parameters.

The search request schema correctly defines required fields (query, matches) and optional fields (contextLines, whole), with clear comments explaining their purpose.

---

85-94: Repository schemas are clear and concise.

Both the repository schema and list repositories response schema are well-structured with all necessary fields.

---

96-105: File source request and response schemas look good.

These schemas appropriately define the fields needed for requesting and receiving file source content, including making the branch field optional in the request schema.

---

107-111: Good standardized error schema.

The service error schema provides a consistent structure for error responses with status code, error code, and message - this will help with client-side error handling.