An MCP server for monkeytype.com API that exposes tools for LLM interaction
MonkeyType MCP exposes 20 tools with generally adequate naming and schemas, but descriptions are inconsistent in quality and some critical parameters lack guidance. Tool names follow the verb_noun pattern well (get_*, check_*, send_*), which is good. However, descriptions range from acceptable (check_username: 37 chars) to minimal (get_tags, get_stats, get_current_test_activity: all under 30 chars), violating the 10 - 1024 character guideline and failing to provide actionable context for LLM selection. Input schemas are present for all tools and use Zod with JSON Schema export, providing type safety. However, output schemas are completely undocumented, the source shows no definition of what these tools return, forcing LLMs to reason blindly about response structure. Parameter descriptions are sparse: mode2 in get_personal_bests says 'Secondary mode parameter for time mode (e.g., 15, 30, 60, 120)' but should constrain to an enum or explicit range. Many optional parameters lack clarity on defaults. Error handling is generic, the callMonkeyTypeApi function returns raw API errors or status codes with no recovery guidance. Security is partially addressed (API key via environment variable, no token in params), but no audit logging is visible. Tool composition is reasonable (single responsibility), but lacks pagination metadata in output schemas and doesn't document response field structure. Baseline expectation is 4 params per tool (p10=1, p90=8); this server averages ~2 params, which is good for simplicity but loses documentation opportunity.
Check if a username is available on MonkeyType
Get server configuration
Get user's current test activity
Get daily typing test leaderboard
Get user's last typing test result
Get typing test leaderboard
Get user's rank on the leaderboard
Nine tools have descriptions under 30 characters (get_tags, get_stats, get_current_test_activity, get_streak, get_typing_stats, get_psas, get_configuration, and others). Examples: 'Get user's tags' (14 chars), 'Get server configuration' (23 chars). They do not explain WHEN to use the tool, WHAT it returns, or dependencies.
No output schemas are documented anywhere in the source code. The callMonkeyTypeApi function returns raw response.data without shape validation or field documentation. LLMs cannot plan downstream calls or extract required fields (e.g., user_id for chaining) without knowing what fields are returned. Rubric requires 100% of A+ tools document return types.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 56 | 2026-07-28+ | v2 |
Get user's personal best typing scores
Get user's profile information
Get public service announcements
Get a specific typing test result by ID
Get user's typing test results
Get speed histogram data
Get user's typing statistics
Get user's typing streak information
Get user's tags
Get global typing statistics
Get weekly XP leaderboard
Check if quote submission is enabled
Send a forgot password email to a user
Parameters mode2 in get_personal_bests and get_leaderboard_rank are described as 'Secondary mode parameter' with examples like '15, 30, 60, 120' or '15, 60, etc.' but are not constrained to an enum. This violates the constrained-input pattern: free-form strings invite hallucinated values. The description should be replaced with enum: ['15', '30', '60', '120'] or similar, and the description updated to explain that these are time durations in seconds.
Error handling is generic. The callMonkeyTypeApi function catches errors and returns either { error: <api response>, statusCode } or { error: <message> }. No recovery guidance is provided. Try search_users() with a partial name.' Current errors give the agent nothing actionable, a 404 on get_profile should suggest using check_username first, or that the username may not exist.
No pagination metadata in tool descriptions or inferred responses. Tools like get_results, get_leaderboard, and get_daily_leaderboard accept limit/offset/skip parameters but do not document total count, has_more, or next_cursor in their descriptions. Rubric states: 'Tools returning lists should accept page/offset and limit parameters and return a total count or next_cursor.' Without this, LLMs cannot reliably iterate through large result sets.
send_forgot_password_email is a WRITE operation that sends an irreversible email. The description does not mention this, and there is no confirmation step or dry-run capability. Agents make mistakes, a confirm_before_execute pattern prevents catastrophic errors.' An LLM could accidentally trigger password resets for many users.
Parameter mode in get_personal_bests is optional with description 'Mode for personal bests (time, words, quote, zen). Defaults to 'time'' but the description does not clarify the enum. It should be: mode: z.enum(['time', 'words', 'quote', 'zen']).optional() with description 'Typing mode (default: time).' Currently it is a free-form string that could accept invalid values.