MCP server for Formula 1 data retrieval and race strategy analysis, leveraging FastF1 library and OpenF1 API
The F1 MCP server presents a domain-specific toolkit with 16 tools spanning FastF1 and OpenF1 APIs. While all tools have descriptions and declared input schemas, quality is inconsistent. Major gaps: (1) Parameter descriptions lack enforcement details, many use union types (string|integer|null) without enums or clear constraints, inviting hallucinated values. (2) Output schemas are almost entirely undocumented, the source shows tool signatures but no declared return structures for LLMs. (3) Error handling is minimal; most tools return error strings inline rather than guiding recovery. (4) Descriptions are present but generic, many under 100 chars without actionable context on when/why to use each tool. (5) No tool annotations (readonly, destructive, idempotent hints). Example: get_session returns a FastF1 Session object, but LLMs cannot know which fields are available or how to chain into downstream tools. The server is functional for domain experts but fragile for autonomous agents, parameter ambiguity and absent output schemas will cause tool selection errors and chaining failures.
Apply one or more filter strings to an API endpoint URL.
Get the championship standing for a specific constructor in a given year.
Get the championship standing for a specific driver in a given year.
Retrieve the API endpoint URL and filter metadata for a given OpenF1 endpoint.
Retrieve a list of all available OpenF1 API endpoints.
Retrieve detailed information about a specific constructor.
Retrieve detailed information about a specific driver.
Output schemas undocumented. No tool declares what fields LLMs should expect in responses. get_session returns a FastF1 Session object with unknown structure; get_season_calendar returns a string; get_session_results returns a DataFrame. LLMs cannot chain outputs without explicit schema documentation.
Union type parameters lack enum constraints and format guidance. session_type accepts 'string|integer|null' with note that it could be 'FP1', 'Q', 'R', 'Sprint', but LLMs will hallucinate invalid values like 'free_practice_1' or 'qualifying_round'. No enum field enforces valid options.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 52 | <=2025-11-25 | v2 |
| 2026-03-09 | D | 58 | - | v1 |
Retrieve detailed information about a specific OpenF1 API endpoint.
Retrieve information about a specific Formula 1 event.
Retrieve detailed information about a specific OpenF1 API filter.
Create a filter string for OpenF1 API requests.
Get the complete race calendar for a specific F1 season.
Retrieve a specific Formula 1 session.
Retrieve and format the results of a specific session.
Send an HTTP GET request to the specified API endpoint and return the JSON response.
Generate a visualization of the track with specified data.
Parameter descriptions lack actionable constraints. 'round (string|integer)' accepts both 'Monaco' and '1' but does not document which inputs work with which endpoints, fallback behavior, or validation rules. driver_name expects 'Full name of the driver (e.g., 'Lewis Hamilton')' but no description of how partial names, abbreviations (HAM), or aliases are handled.
Error handling returns inline strings instead of structured errors with recovery guidance. get_session returns 'Session type X is not available. Supported session types: [...]' as a string, LLMs will treat this as a valid result and attempt to parse it. No error classification (retryable vs. fatal) or suggested next action.
No tool annotations. None of the 16 tools declare readOnlyHint, destructiveHint, or idempotentHint. All tools are read-only (safe to retry), but LLMs cannot infer this, they may assume side effects and plan inefficiently.
Descriptions lack context on tool selection and common use cases. Most descriptions state WHAT the tool does (e.g., 'Retrieve information about a specific Formula 1 event') but not WHEN to call it, how it differs from similar tools, or what prerequisites exist. No hints on discovery or prerequisites.
Missing pagination and result limits for discovery tools. get_api_endpoints and get_api_endpoint return all available endpoints with no limit or pagination, risking context window overflow. No documented cap on result size.
Tool composition gaps in OpenF1 tools. The flow get_api_endpoint → get_filter_string → apply_filters → send_request requires 4 sequential calls for a single API query. No high-level wrapper (e.g., query_openf1_api) combines these steps. Agents will waste tokens and context on multi-step choreography.