This project helps you create and manage a local MCP (Model Context Protocol) server using Composio's 500+ integrations.

1. Composio API Key: Get yours from Composio Platform
2. Auth Configurations: Create auth configs for your toolkits in the Composio Dashboard
3. Python 3.8+: Make sure Python is installed

pip install -r requirements.txt

Copy the example environment file and fill in your credentials:

cp .env.example .env

Edit .env and add:

• Your COMPOSIO_API_KEY
• Your auth config IDs (replace the placeholder IDs in mcp_server.py)
• Optional: API keys for OpenAI/Anthropic if using examples

Before running the server, you need to create auth configurations for your toolkits:

1. Go to Composio Dashboard → Auth Configs
2. Create auth configs for each toolkit you want to use (e.g., Gmail, Google Calendar)
3. Copy the auth config IDs (format: ac_xyz123)
4. Update mcp_server.py with your actual auth config IDs

Run the main script to create your MCP server:

python mcp_server.py

This will:

• Create an MCP server with your configured toolkits
• Generate a unique URL for your user
• Display the MCP server URL to use with AI providers

Important: Users must authenticate with the toolkits before the MCP server can use them.

See Hosted Authentication for how to connect user accounts.

# Set your MCP server URL in .env as MCP_SERVER_URL
python openai_example.py

# Set your MCP server URL in .env as MCP_SERVER_URL
python anthropic_example.py

.
├── mcp_server.py          # Main script to create and manage MCP server
├── openai_example.py      # Example usage with OpenAI
├── anthropic_example.py   # Example usage with Anthropic
├── requirements.txt       # Python dependencies
├── .env.example          # Environment variables template
└── README.md             # This file

Edit mcp_server.py to add or modify toolkits:

toolkits=[
    {
        "toolkit": "gmail",
        "auth_config": "ac_xyz123"  # Your auth config ID
    },
    {
        "toolkit": "googlecalendar",
        "auth_config": "ac_abc456"  # Your auth config ID
    }
]

Modify the allowed_tools list in mcp_server.py:

allowed_tools=[
    "GMAIL_FETCH_EMAILS", 
    "GMAIL_SEND_EMAIL", 
    "GOOGLECALENDAR_EVENTS_LIST"
]

Composio supports 500+ integrations. Check the Composio Docs for the full list of available toolkits and their tools.

• Make sure you've created a .env file with your API key

• Verify your auth config IDs are correct
• Make sure auth configs are created in the Composio dashboard

• Ensure users have authenticated with the toolkits
• Check that the user_id matches the authenticated user

• Composio Documentation
• MCP Server Management
• MCP Providers Guide
• Composio Discord
• GitHub Issues

┌──────────────┐      WebSocket        ┌──────────────────────┐
│   Frontend   │  <------------------> │     FastAPI          │
│ React + D3   │                      │  Memory + Graph API  │
└──────────────┘                       └─────────┬────────────┘
                              Faiss/Sbert  │   Neo4j Bolt
                                            │
                                     ┌──────┴─────┐
                                     │  Storage   │
                                     └────────────┘

• backend/ – FastAPI, Sentence-Transformers + FAISS for semantic memory, Whisper for audio, Neo4j for entity graph, ComposioClient wrapper.
• frontend/ – React + Tailwind + D3, real-time graph via /ws.
• docker-compose orchestrates Neo4j, backend, frontend.

# build & start everything
docker compose up --build

# dashboard
http://localhost:5173      # React UI
http://localhost:7474      # Neo4j Browser
http://localhost:8000/docs # FastAPI docs

Environment variables can be placed in .env (loaded by backend):

COMPOSIO_API_KEY=...
NEO4J_URI=bolt://neo4j:7687
NEO4J_USER=neo4j
NEO4J_PASSWORD=password

Interactive OpenAPI docs served at /docs.

The React app auto-connects to ws://localhost:8000/ws and renders the graph. It expects memories and entities; circles are colour-coded.

Development outside Docker:

cd frontend
npm i
npm run dev

• Tests – add pytest modules under tests/.
• Lint – frontend uses TypeScript strict mode; backend should follow ruff / black (not enforced here).

• Composio MCP servers only support Streamable HTTP transport
• MCP server URLs are unique per user
• Tools requiring authentication won't work until users connect their accounts