X/Twitter data MCP server powered by x402 micropayments. Pay per request in USDC on Base.
twit-mcp demonstrates solid naming conventions and parameter descriptions, but suffers from missing output schema documentation, lack of error guidance, and no tool annotations. All 13 tools are properly registered with Zod schemas and descriptions present in source. Naming follows verb_noun pattern (get_*, search_*). However, output schemas are not documented, tools return JSON.stringify of API responses without explaining what fields agents should expect. Error handling is minimal: the generic `call()` function throws raw API errors with no recovery guidance. No tool annotations (readOnlyHint/destructiveHint) despite all tools being READ_ONLY. Parameter descriptions are functional but could be more specific about formats and constraints. The server achieves moderate quality suitable for internal use but lacks production-grade polish for critical agent deployments.
Retrieve the full content of an X Article by the tweet ID of the post that hosts it. Returns article title, preview text, cover image, publish date, and full article_content as Markdown (headings, bold, italic, lists, images). Not available in the official X API.
Retrieve details of an X Community by its numeric ID. Returns name, description, member count, join policy, topic, rules, banner image, and admin/creator profiles. Not available in the official X API.
Retrieve latest posts from an X Community. Returns ~20 tweets per page with author profiles. Use next_token to paginate. Not available in the official X API.
Retrieve a single tweet/post by its ID. Includes author info, text, metrics, and metadata.
Retrieve multiple tweets/posts by their IDs in a single request (max 50 IDs).
No output schema documentation. Tools return raw JSON from API via JSON.stringify() with no description of returned fields, types, or structure. LLMs cannot plan downstream calls or validate results without knowing the output schema.
Inadequate error handling and recovery guidance. The generic call() function at src/index.ts throws raw API errors like 'API error 404: ...' with no actionable recovery path for the LLM. Errors should guide the agent to retry, try an alternative tool, or ask the user.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 57 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 0 | - | v1 |
Retrieve a Twitter/X user profile by their numeric ID.
Retrieve a Twitter/X user profile by username. Returns id, name, bio, follower count, verified status, and more.
Retrieve the followers of a Twitter/X user. Returns ~50 users per page. Use next_token to paginate.
Retrieve the accounts a Twitter/X user follows. Returns ~50 users per page. Use next_token to paginate.
Retrieve a user's recent tweets/posts. Returns ~20 tweets per page. Use next_token to paginate.
Retrieve multiple Twitter/X user profiles in a single request (max 50 IDs).
Search the full Twitter/X archive with advanced filters. Returns ~20 tweets per page. At least one filter is required.
Search for Twitter/X users by keyword. Returns up to 20 users per page. Use next_token to paginate.
Missing tool annotations. All 13 tools are READ_ONLY but lack explicit readOnlyHint annotations in the tool registration. The MCP 2026-07-28 spec supports tool annotations (readOnlyHint, destructiveHint, idempotentHint) which should be used for clarity.
search_tweets parameter documentation lacks clarity on format constraints. Parameters like 'since' and 'until' are documented as 'Start date inclusive (YYYY-MM-DD)' but Zod schema shows z.string() with no pattern validation. LLMs may pass invalid date formats.
No result limits or pagination guidance in descriptions. Many tools return '~20 tweets per page' or '~50 users per page' but descriptions lack explicit limits and do not warn agents about context window impact when iterating with next_token.
Batch tools (get_users, get_tweets) accept comma-separated IDs as strings rather than arrays. The description warns 'max 50 IDs' but Zod validation does not enforce this limit. An LLM could pass 100+ comma-separated values without validation.