Pull request overview

This PR adds a new get_file_blame tool that enables users to retrieve git blame information for files in GitHub repositories using the GraphQL API. The implementation includes comprehensive test coverage, proper error handling, and complete documentation updates.

Key Changes

• Added GetFileBlame function that queries GitHub's GraphQL API to retrieve blame information showing who last modified each line of a file
• Integrated the new tool into the default toolset alongside other repository management tools
• Created comprehensive unit tests covering successful operations, error scenarios, and edge cases like missing default branches

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated no comments.

Thanks for adding this tool — blame support is a great addition! I have a few concerns I'd love to see addressed:

1. Redundant GraphQL call for default branch: When ref is empty, there's a separate query just to fetch the default branch before the blame query. Please consider using HEAD as the expression or combining both into a single query to avoid the extra round-trip.

2. No pagination or size limit: A blame response for a large file (thousands of lines) could produce a massive JSON payload that blows past MCP response size limits or LLM context windows. Please add some form of truncation or line-range filtering — other tools in this codebase handle pagination, and this one should too.

3. No validation that object resolved to a Commit: If the ref expression resolves to a Tree or Blob, the ... on Commit fragment silently returns zero data and the tool returns an empty ranges array with no error. Let's add a nil/empty check on the commit object with a meaningful error message so users aren't left guessing.

4. No validation on path input: The path parameter is passed directly into the GraphQL query with no sanitization. Please add checks for empty strings, absolute paths, or path traversal patterns (../) — other tools in the codebase validate their path inputs.

5. Inline struct types: BlameRange and BlameResult are defined inside the handler closure. Let's move these to package-level types, consistent with how other tools structure their response types — it makes them testable and reusable.

6. age field is opaque: The age integer from GitHub's GraphQL API represents relative age compared to other ranges, not an absolute value. Please add some documentation in the tool description or response so an LLM consuming this doesn't misinterpret it.

7. Commit message not truncated: Full commit messages (including multi-line bodies) are included for every blame range. For a file with many ranges, this creates enormous redundant output — the same commit message repeated for every range it touches. Please consider including only the first line (subject), or deduplicating commits into a separate section.

8. e2e test changes are unrelated: The e2e/e2e_test.go changes (NewMCPServer → NewStdioMCPServer, adding Version, Logger) look like merge conflict resolution artifacts. Let's clean these up with a rebase so the PR diff only contains blame-related changes.

9. Missing newline at end of snap file: get_file_blame.snap is missing a trailing newline — minor, but please add one to avoid noisy diffs later.

10. No test for empty blame ranges: There's no test case for when the blame query returns successfully but with zero ranges (e.g., empty file). Please add one to verify the response structure is still valid in that case.

Sorry for slow merge, adding it behind feature flag, and in insiders mode so it can be tried by everyone.