Agent MCP

# Agent-MCP Testing Suite Automated testing environment for Agent-MCP with Claude Code integration. ## Quick Start ```bash # Start test environment ./run-test.sh # Clean up when done ./cleanup.sh ``` ## What It Does The testing suite automatically: 1. **Creates isolated test environment** in `tests/test-1/` 2. **Starts MCP server** on available port (default: 8002) 3. **Launches Claude Code** in separate tmux session 4. **Connects Claude to MCP** server via SSE 5. **Initializes admin agent** with extracted admin token 6. **Drops you into Claude session** ready to test ## Files Created ``` testing-suite/ ├── tests/ │ └── test-1/ │ ├── project/ # Test project directory │ │ ├── README.md │ │ ├── pyproject.toml │ │ └── src/ │ └── test-info.txt # Test environment details ├── run-test.sh # Main test script ├── cleanup.sh # Cleanup script └── README.md # This file ``` ## Usage ### Basic Usage ```bash ./run-test.sh ``` ### With Custom Project Directory ```bash ./run-test.sh --project-dir /path/to/your/project ``` ### With Custom Port ```bash ./run-test.sh --port 8003 ``` ## Tmux Sessions The script creates two tmux sessions: - **`agentmcp-test`** - MCP server running - **`claude-test`** - Claude Code interface ### Accessing Sessions ```bash # Attach to Claude (main testing interface) tmux attach-session -t claude-test # View MCP server logs tmux attach-session -t agentmcp-test # Send commands to Claude programmatically tmux send-keys -t claude-test 'your message here' Enter ``` ## Testing Workflow 1. **Start environment**: `./run-test.sh` 2. **Test features** in Claude session (auto-attached) 3. **View server logs** in MCP session if needed 4. **Clean up**: `./cleanup.sh` when done ## Admin Agent The admin agent is automatically initialized with: - **Admin token** extracted from server - **Coordination role** for managing worker agents - **RAG access** for project context - **Task management** capabilities Ready to: - Create worker agents - Assign tasks - Coordinate development work - Test the new testing agent auto-launch feature ## Cleanup Options ```bash # Clean up sessions only (preserve files) ./cleanup.sh # Clean up everything including test files ./cleanup.sh --remove-files ``` ## Dependencies - `uv` - Python package manager - `tmux` - Terminal multiplexer - `claude` - Claude Code CLI - `curl` - For server health checks ## Troubleshooting ### Server Won't Start - Check if port is already in use - Verify `uv` and dependencies are installed - Check MCP server logs: `tmux capture-pane -t agentmcp-test -p` ### Claude Connection Issues - Ensure MCP server is running first - Check Claude MCP configuration: `claude mcp list` - Verify SSE endpoint: `curl http://localhost:8002/sse` ### Token Extraction Failed - MCP server may not have started properly - Check server output for admin token display - Look in tmux session: `tmux capture-pane -t agentmcp-test -p` ## Testing the Testing Agent Feature Once the environment is running, you can test the new auto-testing feature: 1. Create a task: `create_task` 2. Create a worker agent and assign the task 3. Have the worker complete the task (status → "completed") 4. Watch as a testing agent automatically launches 5. Observe the testing agent's critical validation process The testing agent will: - Pause the completing agent - Launch in its own tmux session - Get enriched context about what was completed - Perform heavy criticism and real functionality tests - Send pass/fail feedback to the original agent