Agent Assistant is a system that allows AI agents to interact with human users through a web or mobile interface. It consists of four main components:

1. agentassistant-srv: The main server that handles RPC requests and serves the web interface
2. agentassistant-mcp: An MCP (Model Context Protocol) server that provides tools for AI agents
3. Web Interface: A modern React-based web application for user interaction
4. Flutter Client: A mobile and desktop application for real-time interaction

Note: There is an interreactiveui-askquetion branch, which is like an open-code question tool. You can just select the answer; no need for typing if the answer is provided by the agent.

AI Agent (Claude, etc.)
    ↓ (MCP Protocol)
agentassistant-mcp
    ↓ (Connect-Go RPC)
agentassistant-srv
    ↙ (WebSocket) ↘
Web Interface   Flutter Client

• MCP Integration: Provides ask_question and work_report tools for AI agents
• Real-time Communication: WebSocket-based communication between server and clients
• Modern Web UI: React-based interface with Shadcn/ui components
• Cross-platform Mobile/Desktop: Flutter-based application for Android, iOS, Linux, Windows, and macOS
• Token-based Authentication: Simple token-based authentication system
• Cross-platform Server: Works on Linux, macOS, and Windows

# Build the server
go build -o agentassistant-srv ./cmd/agentassistant-srv

# Build the MCP server
go build -o agentassistant-mcp ./cmd/agentassistant-mcp

./agentassistant-srv

The server will start on port 8080 and serve:

• Web interface at http://localhost:8080
• WebSocket endpoint at ws://localhost:8080/ws
• RPC endpoints for Connect-Go

Edit agentassistant-mcp.toml:

agentassistant_server_host = "127.0.0.1"
agentassistant_server_port = 8080
agentassistant_server_token = "test-token"

Start the MCP server:

./agentassistant-mcp

Open your browser and go to:

http://localhost:8080?token=test-token

The web interface is a React application located in the web/ directory.

cd web
npm install
npm run build

The built files will be placed in web/dist/ and automatically served by the server.

For development:

cd web
npm run dev

The Flutter client is located in the flutterclient/ directory.

Prerequisites:

• Flutter SDK (>= 3.32.0)
• Protocol Buffers compiler (protoc)
• protoc-gen-dart plugin (dart pub global activate protoc_plugin)

Steps:

1. Install dependencies:

   cd flutterclient
   flutter pub get

2. Generate Protobuf files:

   ./generate_proto.sh
   # or
   protoc --proto_path=../proto --dart_out=lib/proto ../proto/agentassist.proto

3. Run the application:

   flutter run

4. Build for specific platforms:

   flutter build apk --release    # Android
   flutter build ios --release    # iOS
   flutter build linux --release  # Linux

go test ./...

Create agentassistant-mcp.toml:

agentassistant_server_host = "127.0.0.1"
agentassistant_server_port = 8080
agentassistant_server_token = "your-token-here"

MCP Server:

./agentassistant-mcp -host localhost -port 8080 -token your-token -web

• -host: Server host (default: 127.0.0.1)
• -port: Server port (default: 8080)
• -token: Authentication token (default: test-token)
• -web: Open web interface in browser

Ask a question to the user through the web interface.

Parameters:

• project_directory (string): Current project directory
• question (string): Question to ask the user
• timeout (number): Timeout in seconds (default: 600)

Notify the user that a task has been completed and send a work report.

Parameters:

• project_directory (string): Current project directory
• summary (string): Summary of the completed task / work report
• timeout (number): Timeout in seconds (default: 600)

• AskQuestion(AskQuestionRequest) returns (AskQuestionResponse)
• WorkReport(WorkReportRequest) returns (WorkReportResponse)

1. During any process, task, or dialogue, whether it's for inquiries, responses, or completing phased tasks, the corresponding MCP tools must be called:
   • Use the ask_question tool when user feedback is needed.
   • Use the work_report tool when a task is completed.
2. Whenever user feedback is received, if the feedback content is not empty, behavior must be adjusted according to the feedback, and the corresponding tools should be called again if necessary.
3. Only when the user explicitly states "end" or "no further interaction needed" can the calling of MCP tools be stopped, and the process is considered complete.
4. Unless an end instruction is received, all key steps must be confirmed with the user through MCP tools.
5. Before completing a task, the work_report tool must be used to report the task completion status to the user and request confirmation.
6. When calling tools, ensure the correct project_directory parameter is provided to maintain contextual accuracy.

中文提示词

1. **重要**：在任何的任务、对话中，如果你有需要澄清的问题/需求/细节/疑问，必须使用 Agent-Assistant 的 `ask_question` 工具来询问用户,根据用户反馈调整你的行为，继续处理。
2. **重要**：在任何的任务、对话中，在任务/对话完成前，必须使用 Agent-Assistant 的 `work_report` 工具来报告任务完成状态给用户，请求确认，得到用户的反馈后，如果不是明确的结束指令，则必须根据反馈内容调整你的行为，继续处理。
3. **重要**：调用 MCP 工具 Agent-Assistant 时，必须提供正确的 `project_directory` 参数，以确保上下文准确性。

英文提示词

1. **Important**：In any task or conversation, if you have any questions, requirements, or doubts, you must use the `ask_question` tool of Agent-Assistant to ask the user. Based on the user's feedback, adjust your behavior and continue processing.
2. **Important**：In any task or conversation, before the task or conversation is completed, you must use the `work_report` tool of Agent-Assistant to report the task completion status to the user and request confirmation. After receiving the user's feedback, if it is not a clear end instruction, you must adjust your behavior based on the feedback content and continue processing.
3. **Important**：When calling the MCP tool of Agent-Assistant, you must provide the correct `project_directory` parameter to ensure contextual accuracy.

This project is licensed under the MIT License.