Skip to content

SETUP.md

Latest commit

 

History

History
122 lines (95 loc) · 3.33 KB

SETUP.md

File metadata and controls

122 lines (95 loc) · 3.33 KB
Raw

A2AMCP Server Setup Guide

Overview

The A2AMCP (Agent-to-Agent Model Context Protocol) server enables communication between AI agents working on the SplitMind platform. It uses Redis for state management and communicates via MCP's STDIO protocol.

Latest Update: Server now uses modern MCP SDK 1.9.3 with proper @server.list_tools() and @server.call_tool() decorators, eliminating connection hanging issues.

Running the Server

  1. Start the Docker containers:

    docker-compose up -d

  2. Verify containers are running:

    docker ps | grep splitmind

    You should see:

    • splitmind-mcp-server - The MCP server container
    • splitmind-redis - Redis for state storage

  3. Check server logs:

    docker logs splitmind-mcp-server

Configuring Claude Desktop

Add the following to your Claude Desktop MCP configuration file:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "splitmind-a2amcp": {
      "command": "docker",
      "args": [
        "exec",
        "-i",
        "splitmind-mcp-server",
        "python",
        "/app/mcp-server-redis.py"
      ],
      "env": {
        "REDIS_URL": "redis://redis:6379"
      }
    }
  }
}

Available MCP Tools

The server implements 16+ A2AMCP API tools using modern MCP SDK patterns:

Agent Management

  • register_agent - Register an agent for a project
  • unregister_agent - Unregister an agent when done
  • list_active_agents - List all active agents
  • heartbeat - Keep agent alive

Todo Management

  • add_todo - Add a todo item
  • update_todo - Update todo status
  • get_my_todos - Get agent's todos

Communication

  • query_agent - Send query to another agent
  • check_messages - Check and retrieve messages
  • respond_to_query - Respond to a specific query

File Coordination

  • announce_file_change - Lock a file before editing
  • release_file_lock - Release file lock after editing
  • get_recent_changes - Get recent file changes

Shared Definitions

  • register_interface - Share a type/interface definition
  • query_interface - Get shared interface definition
  • list_interfaces - List all shared interfaces

Task Completion

  • mark_task_completed - Signal task completion to orchestrator

Technical Implementation:

  • Uses @server.list_tools() to register available tools
  • Uses @server.call_tool() to handle tool execution
  • Proper A2AMCP response format with status, message, and data
  • No connection hanging issues with modern MCP SDK 1.9.3

Testing

Run the test script to verify all endpoints:

python test_endpoints.py

Ports

  • 5050 - MCP server (mapped from internal 5000)
  • 6379 - Redis
  • 8081 - Redis Commander UI (optional, with --profile debug)

Troubleshooting

  1. Server keeps restarting:

    • This is normal for STDIO-based MCP servers
    • The server only runs when Claude connects to it

  2. Connection issues:

    • Ensure Docker is running
    • Check Redis is healthy: docker exec splitmind-redis redis-cli ping
    • Verify port 5050 is not in use

  3. View Redis data:

    docker-compose --profile debug up -d

    Then open http://localhost:8081 for Redis Commander UI