A2AMCP - Agent-to-Agent Model Context Protocol
Enabling Seamless Multi-Agent Collaboration for AI-Powered Development
A2AMCP brings Google's Agent-to-Agent (A2A) communication concepts to the Model Context Protocol (MCP) ecosystem, enabling AI agents to communicate, coordinate, and collaborate in real-time while working on parallel development tasks.
Originally created for SplitMind, A2AMCP solves the critical problem of isolated AI agents working on the same codebase without awareness of each other's changes.
β Server Status: WORKING! All 17 tools implemented and tested. Uses modern MCP SDK 1.9.3.
π Quick Start
Using Docker (Recommended)
# Clone the repository
git clone https://github.com/webdevtodayjason/A2AMCP
cd A2AMCP
# Start the server
docker-compose up -d
# Verify it's running
docker ps | grep splitmind
# Test the connection
python verify_mcp.py
Configure Your Agents
Claude Code (CLI)
# Add the MCP server using Claude Code CLI
claude mcp add splitmind-a2amcp \
-e REDIS_URL=redis://localhost:6379 \
-- docker exec -i splitmind-mcp-server python /app/mcp-server-redis.py
Claude Desktop
Add to your configuration file (~/Library/Application Support/Claude/claude_desktop_config.json
on macOS):
{
"mcpServers": {
"splitmind-a2amcp": {
"command": "docker",
"args": ["exec", "-i", "splitmind-mcp-server", "python", "/app/mcp-server-redis.py"],
"env": {
"REDIS_URL": "redis://redis:6379"
}
}
}
}
π― What Problem Does A2AMCP Solve?
When multiple AI agents work on the same codebase:
- Without A2AMCP: Agents create conflicting code, duplicate efforts, and cause merge conflicts
- With A2AMCP: Agents coordinate, share interfaces, prevent conflicts, and work as a team
Generic Use Cases Beyond SplitMind
A2AMCP can coordinate any multi-agent scenario:
- Microservices: Different agents building separate services
- Full-Stack Apps: Frontend and backend agents collaborating
- Documentation: Multiple agents creating interconnected docs
- Testing: Test writers coordinating with feature developers
- Refactoring: Agents working on different modules simultaneously
ποΈ Architecture
βββββββββββββββββββ
β A2AMCP Server β β Persistent Redis-backed MCP server
β (Port 5050) β handling all agent communication
ββββββββββ¬βββββββββ
β STDIO Protocol (MCP)
ββββββ΄βββββ¬ββββββββββ¬ββββββββββ
βΌ βΌ βΌ βΌ
ββββββββββββββββββββββββββββββββββββββββ
βAgent 1 ββAgent 2 ββAgent 3 ββAgent N β
βAuth ββProfile ββAPI ββFrontendβ
ββββββββββββββββββββββββββββββββββββββββ
π§ Core Features
1. Real-time Agent Communication
- Direct queries between agents
- Broadcast messaging
- Async message queues
2. File Conflict Prevention
- Automatic file locking
- Conflict detection
- Negotiation strategies
3. Shared Context Management
- Interface/type registry
- API contract sharing
- Dependency tracking
4. Task Transparency
- Todo list management
- Progress visibility
- Completion tracking
- Task completion signaling
5. Multi-Project Support
- Isolated project namespaces
- Redis-backed persistence
- Automatic cleanup
6. Modern MCP Integration
- Uses MCP SDK 1.9.3 with proper decorators
@server.list_tools()
and@server.call_tool()
patterns- STDIO-based communication protocol
- Full A2AMCP API compliance with 17 tools implemented
π¦ Installation Options
Docker Compose (Production)
services:
mcp-server:
build: .
container_name: splitmind-mcp-server
ports:
- "5050:5000" # Changed from 5000 to avoid conflicts
environment:
- REDIS_URL=redis://redis:6379
- LOG_LEVEL=INFO
depends_on:
redis:
condition: service_healthy
restart: unless-stopped
redis:
image: redis:7-alpine
container_name: splitmind-redis
ports:
- "6379:6379"
volumes:
- redis-data:/data
healthcheck:
test: ["CMD", "redis-cli", "ping"]
interval: 10s
timeout: 5s
retries: 5
volumes:
redis-data:
driver: local
Python SDK
pip install a2amcp-sdk
JavaScript/TypeScript SDK (Coming Soon)
npm install @a2amcp/sdk
π¦ Usage Example
Python SDK
from a2amcp import A2AMCPClient, Project, Agent
async def run_agent():
client = A2AMCPClient("localhost:5000")
project = Project(client, "my-app")
async with Agent(project, "001", "feature/auth", "Build authentication") as agent:
# Agent automatically registers and maintains heartbeat
# Coordinate file access
async with agent.files.coordinate("src/models/user.ts") as file:
# File is locked, safe to modify
pass
# File automatically released
# Share interfaces
await project.interfaces.register(
agent.session_name,
"User",
"interface User { id: string; email: string; }"
)
Direct MCP Tool Usage
# Register agent
register_agent("my-project", "task-001", "001", "feature/auth", "Building authentication")
# Query another agent
query_agent("my-project", "task-001", "task-002", "interface", "What's the User schema?")
# Share interface
register_interface("my-project", "task-001", "User", "interface User {...}")
π Documentation
- Claude Code Setup Guide
- Installation & Setup
- Full API Reference
- Python SDK Documentation
- Architecture Overview
- SplitMind Integration Guide
π οΈ SDKs and Tools
Available Now
- Python SDK: Full-featured SDK with async support
- Docker Deployment: Production-ready containers
In Development
- JavaScript/TypeScript SDK: For Node.js and browser
- CLI Tools: Command-line interface for monitoring
- Go SDK: High-performance orchestration
- Testing Framework: Mock servers and test utilities
See SDK Development Progress for details.
π€ Integration with AI Frameworks
A2AMCP is designed to work with:
- SplitMind - Original use case
- Claude Code (via MCP)
- Any MCP-compatible AI agent
- Future: LangChain, CrewAI, AutoGen
π How It Differs from A2A
While inspired by Google's A2A protocol, A2AMCP makes specific design choices for AI code development:
Feature | Google A2A | A2AMCP |
---|---|---|
Protocol | HTTP-based | MCP tools |
State | Stateless | Redis persistence |
Focus | Generic tasks | Code development |
Deployment | Per-agent servers | Single shared server |
π Roadmap
- Core MCP server with Redis
- Modern MCP SDK 1.9.3 integration
- Fixed decorator patterns (
@server.list_tools()
,@server.call_tool()
) - Python SDK
- Docker deployment
- All 17 A2AMCP API tools implemented and tested
- Health check endpoint for monitoring
- Verification script for testing connectivity
- JavaScript/TypeScript SDK
- CLI monitoring tools
- SplitMind native integration
- Framework adapters (LangChain, CrewAI)
- Enterprise features
π οΈ Troubleshooting
Agents can't see mcp__splitmind-a2amcp__
tools
- Restart Claude Desktop - MCP connections are established at startup
- Verify server is running:
docker ps | grep splitmind
- Check health endpoint:
curl http://localhost:5050/health
- Run verification script:
python verify_mcp.py
- Check configuration: Ensure
~/Library/Application Support/Claude/claude_desktop_config.json
contains the A2AMCP server configuration
Common Issues
- "Tool 'X' not yet implemented" - Fixed in latest version, pull latest changes
- Connection failed - Ensure Docker is running and ports 5050/6379 are free
- Redis connection errors - Wait for Redis to be ready (takes ~5-10 seconds on startup)
π€ Contributing
We welcome contributions! See CONTRIBUTING.md for guidelines.
Development Setup
# Clone repository
git clone https://github.com/webdevtodayjason/A2AMCP
cd A2AMCP
# Install dependencies
pip install -r requirements.txt
# Run tests
pytest
# Start development server
docker-compose -f docker-compose.dev.yml up
π Performance
- Handles 100+ concurrent agents
- Sub-second message delivery
- Automatic cleanup of dead agents
- Horizontal scaling ready
π Security
- Project isolation
- Optional authentication (coming soon)
- Encrypted communication (roadmap)
- Audit logging
π License
MIT License - see LICENSE file.
π Acknowledgments
- Inspired by Google's A2A Protocol
- Built for SplitMind
- Powered by Model Context Protocol
π Support
- Issues: GitHub Issues
- Discussions: GitHub Discussions
- Discord: Coming soon
A2AMCP - Turning isolated AI agents into coordinated development teams