twitterapi-io-mcp

# Recipe: Typo-Tolerant Search (“twet object”) `search_twitterapi_docs` is typo-tolerant (tokenization + n-gram fuzzy scoring). No special flags required. ## JSON tool call ```json { "tool": "search_twitterapi_docs", "arguments": { "query": "twet object", "max_results": 10 } } ``` ## Result types (how to chain) - If `type === "endpoint"`: use `result.name` as `endpoint_name` for `get_twitterapi_endpoint`. - If `type === "page"`: use `result.name` as `guide_name` for `get_twitterapi_guide` (or fetch `result.url` via `get_twitterapi_url`). - If `type === "blog"`: fetch `result.url` via `get_twitterapi_url`. ## Practical pattern (search → endpoint details) ```js async function tool(name, args) { const res = await callTool(name, args); if (res?.isError) throw new Error(res?.content?.[0]?.text ?? `Tool failed: ${name}`); return res?.structuredContent ?? {}; } const s = await tool("search_twitterapi_docs", { query: "twet object", max_results: 10 }); // Prefer an endpoint hit when you want API shapes/fields. const endpoints = (s.results ?? []).filter((r) => r.type === "endpoint"); const ranked = endpoints.length ? endpoints : (s.results ?? []); if (!ranked.length) throw new Error("No results for typo query"); // Ambiguity handling: if the top-2 are close, ask the user (or refine the query). ranked.sort((a, b) => (b.score ?? 0) - (a.score ?? 0)); const [best, second] = ranked; if (second && Math.abs((best.score ?? 0) - (second.score ?? 0)) < 2) { // e.g. present [best.name, second.name] and ask which one is intended } // If the best hit isn't an endpoint, keep it as a doc result; otherwise fetch endpoint details. if (best.type === "endpoint" && best.name) { const doc = await tool("get_twitterapi_endpoint", { endpoint_name: best.name }); } ``` ## Tuning & ambiguity - If the query is very short/typo-heavy, try `max_results: 15–20` and filter by `type`. - Add 1 context token to stabilize matches: `"twet object tweet"`, `"twet object fields"`. - If results are empty, retry with a corrected query: `"tweet object"`, `"tweet objects"`. - If multiple hits look plausible, compare `score` and prefer the higher-scoring `type: "endpoint"`.