FAQ

Frequently asked questions about the Braking Lab MCP Server.

General

What is the MCP Server?

The MCP Server is a cloud service that connects AI assistants to your Braking Lab data. It uses the Model Context Protocol (MCP) — an open standard for AI-to-service communication. Any compatible AI client can connect and query your telemetry, generate coaching reports, and manage your race calendar.

What AI clients can I use?

Any client that supports the Model Context Protocol:

  • Claude Desktop — Full support with streamable HTTP
  • ChatGPT — Via MCP plugin support
  • Cursor — Native MCP integration
  • Windsurf — Native MCP integration
  • Gemini — Via MCP support
  • Custom clients — Any client implementing the MCP specification

Do I need to be a developer?

No. The AI writes all the code for you. You just ask questions in plain language like “Where am I losing time at Spa?” or “Generate a strategy for my next race.” The code-mode sandbox is invisible to you.

Is it free?

The MCP Server is included free with your Braking Lab account. You only pay for the AI model through your chosen client (e.g., Claude Desktop subscription, ChatGPT Plus). Braking Lab itself is free.

Privacy & Security

Is my data private?

Yes. The MCP Server only accesses data belonging to your authenticated account. Every query verifies data ownership before returning results. Your data is never shared with other users or used for AI training.

Can the AI access other users’ data?

No. Ownership checks are enforced on every function call. Even if an AI tries to pass another user’s session ID, the request will be rejected.

Is the code the AI writes stored?

No. Code is transpiled, executed in a sandbox, and discarded. We don’t log or store the TypeScript that AI generates.

What happens if the AI writes malicious code?

The sandbox has no network access, no filesystem access, and a 10-second timeout. The worst that can happen is the execution times out or returns an error. It cannot access external services, modify server state outside approved functions, or exfiltrate data.

Technical

How many functions are available?

Currently 55 functions across 11 domain groups covering sessions, stints, corners, braking, coaching, races, notepads, analysis, strategy, iRacing, vault, and more. See the Tool Catalog for the complete list.

What’s the difference between the MCP Server and the Race Engineer?

They’re the same thing viewed from different angles:

  • Race Engineer is the user-facing coaching experience — what you get when you chat with an AI about your driving
  • MCP Server is the technical platform — the API, authentication, and sandbox that makes it possible

The Race Engineer page focuses on what you can do. The MCP Server docs focus on how it works.

Can I call multiple functions in one execution?

Yes. That’s one of the advantages of code-mode — the AI can compose multiple function calls in a single execute_code invocation, up to 20 API calls per execution.

What if my token expires?

AI clients handle token refresh automatically using the refresh token. You shouldn’t need to re-authorize unless you explicitly revoke access.

Is there a rate limit?

The server includes rate limiting on REST API endpoints. The MCP endpoint uses the sandbox constraints (10s timeout, 20 calls per execution) as natural throttling.