Node.js mcp server for iRacing data api
The server defines 6 tools with explicit Zod schemas and basic error handling. However, significant gaps undermine quality: (1) Most tool descriptions are under 20 characters, failing the baseline of 194 chars average and the 20-char minimum threshold. (2) Input parameters lack descriptions entirely, e.g., 'member_ids' array has a description in the schema, but 'driver_name', 'team_id', 'member_id', 'year', 'season' lack meaningful context. (3) Output schemas are undocumented, tool responses return JSON.stringify() of API data with no specification of what fields the LLM should expect. (4) Tool naming is partially clear (verb_noun pattern used, e.g., 'get_members_profile'), but 'driver_lookup' lacks clarity on whether it searches by name, alias, or ID. (5) No enumerated parameters, no pagination support, no output field documentation. (6) Error messages provide minimal recovery guidance, generic 'Failed to get X' strings without hints on what to try next. Per-tool average: 42/100.
Tool to lookup driver data
Tool to get member career data
Tool to get member recap data
Define a tool to get member profile
Define a tool to get team data
Define a tool to login to iRacing
Tool descriptions are critically short (10 - 35 chars), well below the 194-char baseline and failing the 20-char minimum threshold for meaningful context. Examples: 'Define a tool to login to iRacing' (33 chars), 'Define a tool to get team data' (30 chars), 'Tool to lookup driver data' (25 chars). LLMs cannot determine when to use these tools without expanded guidance on what data they return, prerequisites, and related tools.
Input parameters lack descriptions entirely. 'member_ids', 'driver_name', 'team_id', 'member_id', 'year', 'season' are defined in Zod schemas but have no human-readable guidance in tool registration. LLMs cannot infer meaning from names alone, e.g., 'member_id' could be a username, numeric ID, email, or account handle. Parameter descriptions are required for every input.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 42 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 42 | - | v1 |
Output schemas are completely undocumented. Tools return JSON.stringify() of API responses with no specification of what fields exist, types, or structure. LLMs cannot plan downstream tool calls or extract required fields without knowing what the response contains. All tools must document their return type with example fields and types.
Error messages provide no recovery guidance. All tools return generic strings like 'Failed to get member profile: {error}' without suggesting what the agent should try next. Errors must be categorized (retryable vs. user-fixable) and include actionable recovery steps, e.g., 'Member ID not found. Try using driver_lookup() to search by name.'
'driver_lookup' naming is ambiguous. Does it search by driver name, username, member ID, or car number? The description 'Tool to lookup driver data' offers no clarity. Rename to 'search_drivers_by_name' or 'find_driver_by_name' to disambiguate from potential ID-based lookups.
No pagination support. Tools like 'get_members_profile' accept arrays of member IDs but return all results without limit, offset, or next_cursor. Large result sets bloat the context window. Implement page/limit parameters and return total_count or next_cursor.
Parameters lack enums and validation hints. 'year' and 'season' accept any number; no constraints on valid ranges (e.g., year must be 2000 - 2026, season 1 - 4). Enums and min/max constraints prevent hallucinated invalid values.