mcp

This MCP server acts as a bridge between Claude (or other AI assistants) and AnswerRocket's Max AI platform. It dynamically registers tools and skills for a given copilot ID and exposes them to your AI conversations via a standardized MCP interface. In local mode, the server talks directly to a Max instance using an API URL and token. In remote mode, it uses OAuth for authentication and can multi-tenant serve multiple Max instances based on incoming requests. Tools available to clients are determined by the copilot’s skills within Skill Studio and are surfaced through the MCP endpoint so your AI assistant can invoke Max capabilities in-context.

Prerequisites:

• Python 3.10+ and Git
• A Max instance available for development or production use
• uv (as installation/development helper)

Installation steps:

1. Clone the repository

git clone https://github.com/answerrocket/mcp-server.git
cd mcp-server

2. Install uv and dependencies (development workflow)

uv sync

3. Configure environment variables for your chosen mode (local or remote) Local mode (direct API access):

export MCP_MODE=local
export AR_URL=http://your-max-instance:port
export AR_TOKEN=your-sdk-token
export COPILOT_ID=your-copilot-uuid
export MCP_TRANSPORT=stdio

Remote mode (OAuth):

export MCP_MODE=remote
export MCP_TRANSPORT=streamable-http
export MCP_HOST=0.0.0.0
export MCP_PORT=9090

4. Run the server (example using uv with Python entrypoint)

uv run python mcp_server/__main__.py

5. Test the MCP endpoint via your client (e.g., local tooling or an integration in Claude Desktop).

Note: The README also documents alternative deployment paths (e.g., using npx for MCP Inspector or running via Docker in other setups). Adjust path and environment values to your environment.

Tips and common issues:

• In remote mode, MCP_HOST and MCP_PORT define where the server listens; ensure firewall rules allow external access if needed.
• In local mode, ensure AR_URL, AR_TOKEN, and COPILOT_ID are accurate to register the correct copilot tools.
• When using OAuth, tokens are validated against the request domain and the server may need multi-tenancy support; ensure your OAuth server is configured accordingly.
• For development, using the MCP Inspector can help validate tool registrations and responses; if you encounter issues with inspector version mismatches, pin to a known-good version as shown in the README.
• The available tools/skills are dynamic and depend on the copilot’s configuration in Skill Studio; you may not see all Max capabilities until the copilot is correctly registered and has skills published.