A Model Context Protocol server that provides tools to interact with X.com (Twitter) API, including getting user profiles, posting tweets, and searching for tweets.
Three tools are defined with basic schema structures using Zod, but significant quality gaps exist across naming, descriptions, and parameter documentation. Tool names follow verb_noun convention (getUserProfile, postTweet, searchTweets), which is positive. However, descriptions are generic and under-optimized for LLM reasoning. Input schemas are present but parameter descriptions are minimal or missing in the actual implementation. The searchTweets response is malformed (returns raw objects instead of structured text content). No error handling guidance is provided to help LLMs understand recovery options. Overall, definitions are functional but below production quality standards.
Get a user's profile information on twitter or x.com
Post a new tweet on x.com
Search for recent tweets on x.com
searchTweets response is malformed: returns raw objects instead of properly structured MCP content with type='text'. Line `return { content: [tweets, users] };` violates MCP response contract.
postTweet lacks error recovery guidance. Tool is WRITE (destructive) but has no documentation of what can fail (rate limits, auth, moderation) or how LLM should recover. Missing: confirmation pattern for destructive operations.
Parameter descriptions are inadequate. 'query' and 'count' in searchTweets lack explanation of valid formats, constraints beyond numbers, or behavior on edge cases. No guidance on what 'recent' means or pagination strategy.
Output schemas are not documented for any tool. LLMs cannot infer what fields to expect from getUserProfile (does it return metrics? description? profile image URL?) or how to chain results into downstream tools.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 48 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 49 | - | v1 |
No pagination support in searchTweets despite accepting 'count' parameter. Large result sets will blow context window. Missing: next_cursor, total_count, offset/limit pattern.
Tool descriptions are generic and under 100 chars. Baseline for A+ tools is 50-200 chars with WHEN/WHY guidance. Current descriptions lack guidance on tool selection (when to use getUserProfile vs searchTweets?) and prerequisites.
X API requires authentication (Bearer token passed via authHeader). No parameter-level guidance on what happens if token is missing, expired, or invalid. Error handling returns generic 'Error fetching user profile: {error.message}' with no recovery path.