fetchapiMCP · API DOCS

Live social data for agents. One key.

An MCP server and an HTTP API for X/Twitter, Instagram and Telegram data. Each endpoint below was called for real. You see what it returned, what one item is, and which fields matter before you spend a call. Calls run on our provider accounts and are metered to your key. Agents can read /mcp.md or /llms.txt.

Connect


    

MCP endpoint : Streamable HTTP, stateless JSON-RPC. You can search endpoints and read the docs without a key. call_endpoint needs your key as Authorization: Bearer gw_…. Sign in to get a key (free; each key has a rate limit and a monthly $ cap).

How an agent uses it

1 · search_endpoints

Ask for the job, not the vendor: "instagram followers", "tweets of a user". You get tool ids with price per call and required params; every one passed a live test.

2 · get_endpoint

Parameters with working example values, what one item is, the fields worth knowing, and the real answer we got.

3 · call_endpoint

Live data with your parameters. It passes the same checks as the HTTP API: key, scope, rate limit, monthly and daily caps, metering.

Tools

Same thing over HTTP


    

Successful responses look like {"status":"success","data":<the provider's JSON, unchanged>,"meta":{"provider","path","cost_usd","time_ms"}}. Errors are typed: unauthorized, forbidden, rate_limited, budget_exceeded, upstream_error. Add ?raw=1 to get the provider's body as is.

Standard endpoints

One URL per job, the cheapest working provider behind it, one answer shape: /api/v1/google/search?q=, /api/v1/x/user?username=, /api/v1/x/search?q=, /api/v1/tiktok/{user,posts,video,comments,search,hashtag,followers}, /api/v1/instagram/{user,posts,post,comments,followers,search,hashtag}, /api/v1/youtube/video?id=. If a provider fails we try the next; you pay only for the call that answered. MCP: get_data. Docs.

Best price for a job

GET /api/prices?q=search tweets (or the MCP tool best_price, no key needed): the endpoints live here that do it, cheapest first with the exact call, and every provider we've priced across the market with the source of each price. Try it · Docs.

Need something else?

Request an endpoint we don't have yet: POST /api/requests {"want": "..."}, the MCP tool request_endpoint, or the form on the catalogue. We find the cheapest source, its price and licence, and post the answer at the request's status_url (or GET /api/requests with your key). Docs.

Spend less

Every answer carries its upstream cost in meta.cost_usd. Across all keys, 0.4% of calls made 28% of the bill and repeat requests made 18%. The fixes:

  1. Ask for the rows you will read. Per-row prices dominate: set limit / depth (LLM Mentions: limit: 1000 up to $1.10, limit: 10 $0.11).
  2. Use the summary endpoint for counts (e.g. llm_mentions/aggregated_metrics, about $0.10, no row charge).
  3. Cache answers and reuse them while the data can't have changed. A repeated request costs the same again.
  4. Queue what can wait: DataForSEO's standard queue is $0.0006 a Google results page, live is $0.002.
  5. Pick the cheapest endpoint for the job: search_endpoints and compare usd_per_call; the same job can differ 100x.
  6. Retry only timeouts and 5xx. Failed calls are free, but a bad input fails again.

Your own breakdown, with cheaper same-job routes: /account (or the get_usage tool). Guide for agents: /docs/spend-less.md. The long version: How to spend less on data API calls.

Endpoints

Each endpoint was sampled with one real call. Sample answers keep the first 3 items of each list. We list write actions, and endpoints that need the end user's own X login, but never call them. Upstream price is the provider's per-request price on its entry plan ().