mcp

The QuantConnect MCP Server acts as a bridge that lets AI agents (such as Claude or OpenAI o3 Pro) interact with QuantConnect’s cloud platform. Through the MCP, the AI can perform actions like updating projects, writing strategies, running backtests, and deploying live strategies to production. The official server is provided as a Docker image for cross‑platform deployment and to help ensure secure handling of API tokens. To connect an MCP client, configure the MCP server entry in your client’s settings so that requests are forwarded to the Docker container running quantconnect/mcp-server. The server exposes a wide array of tools that map to QuantConnect operations, such as creating projects, editing files, backtesting, and managing live deployments.

Prerequisites:

• Docker Desktop installed on your machine (Linux/macOS/Windows supported).
• Access to QuantConnect MCP Server image on Docker Hub (quantconnect/mcp-server).

Installation steps:

1. Ensure Docker is running and pull the MCP server image (ARM platforms may require the platform flag):

docker pull quantconnect/mcp-server

2. Run the MCP server container with the necessary environment variables. Replace placeholder values with your credentials and preferred agent name:

docker run -i --rm \
  -e QUANTCONNECT_USER_ID=<your_user_id> \
  -e QUANTCONNECT_API_TOKEN=<your_api_token> \
  -e AGENT_NAME="MCP Server" \
  --platform <your_platform> \
  quantconnect/mcp-server

3. If you need to connect via a client like Claude Desktop, update the client config (for Claude Desktop, e.g., claude_desktop_config.json) to point to the MCP server using the docker command and environment variables shown in the example. The configuration typically resides under mcpServers.quantconnect. See the integration guide for details.

4. Verify connectivity and start using: the MCP will expose the available tools and operations once the container is running.

Notes and tips:

• The MCP server is multi-platform capable via Docker. If you are on ARM hardware (e.g., Apple Silicon), you may need to specify --platform linux/arm64 when pulling/running the image.
• Use a unique AGENT_NAME if you run multiple agents to distinguish requests in logs and dashboards.
• The environment variables QUANTCONNECT_USER_ID and QUANTCONNECT_API_TOKEN must be kept secret and won’t be logged by default.
• The server exposes a long list of tools (see the Available Tools section in the README) for reading, creating, updating, and deleting projects, files, backtests, optimizations, live algorithms, object storage, and more. If a tool is not available in your plan, you may need to enable it in your QuantConnect account or check the latest MCP server version.
• To stay up-to-date, periodically pull the latest quantconnect/mcp-server image and restart the container.