Documentation · MCP server
Ask for data in plain language
MCP is a common way to plug a tool into an AI assistant, so the assistant can go and do the job itself instead of telling you how. Add FastFetchz once, and your assistant can pull social and web data for you whenever you ask.
FastFetchz runs a standard remote MCP server with OAuth. Connect it to whichever AI assistant or coding agent you already use and ask for social and web data the way you'd ask a colleague — "Get the Instagram profile for @nasa" is a complete instruction. The assistant picks the right service, runs it against your account and hands back the result.
The connection is authorised in your browser with OAuth, so there is no key to paste and no key to leak. You approve it once on a FastFetchz consent screen, it uses your existing credit balance, and you can revoke it whenever you like.
Server URL — the one thing you paste
https://fastfetchz.com/mcpThis is the only value you ever need. Anything that asks for an API key is not this connector — the REST API uses keys, the MCP server uses OAuth.
Connect your client
Pick the assistant you use. The steps for every other client are one click away, unchanged.
Connect Claude (web)
- 1Go to claude.ai/customize/connectors
- 2Click +, then Add custom connector
- 3Paste https://fastfetchz.com/mcp
- 4Click Add, then Connect, and approve on the FastFetchz consent screen
- 5Enable FastFetchz from the tools menu next to the prompt box, then ask for something — “Get the Instagram profile for @nasa”
Team and Enterprise plans: an organisation owner adds it under Organization settings → Connectors → Add → Custom. After that, each member connects it from their own connectors page and authorises their own FastFetchz account.
Which clients work
We have connected FastFetchz ourselves and made live billed calls from four clients — Claude, ChatGPT, Grok and Cursor — so those are marked tested and working. Because this is a standard remote MCP server (Streamable HTTP with OAuth 2.1), the remaining clients below connect the same way and are marked supported on the strength of their own documented remote MCP support. Client names are their owners' trademarks and imply no partnership.
The five tools
Once connected, your assistant sees five tools. Only fetch costs anything. The other four exist so an assistant can look before it leaps — discover what's available, check the target format, and price the job before spending a single credit.
Lists every platform in the catalog, with how many services each has.
Takes: Nothing.
Every service, what to pass as the target, and what comes back.
Takes: Optionally a platform, to narrow the list.
Calls, credits and dollars for a job before you run it.
Takes: A service and how many items or pages you want.
Runs a data request and returns the same JSON envelope as REST, with credits charged and credits remaining.
Takes: A service and its parameters — usually a target such as a handle or a URL.
Your credit balance and account status.
Takes: Nothing.
What it feels like
Prompts a customer can paste as-is once the connector is live:
- “What FastFetchz services are available for TikTok?”
- “How much would it cost to get 50,000 followers of @nike on X?”
- “Get the Instagram profile for @nasa”
- “Pull the last 20 tweets from @elonmusk and summarise the themes”
- “What's my FastFetchz credit balance?”
- “List every YouTube service and tell me which one needs a video ID”
Before and after
You ask: “How much would it cost to get 50,000 followers of @nike on X?”
Claude does: calls estimate_cost for the X followers service, reports that 50,000 followers is roughly 250 calls at 3 credits each — about 750 credits — and then stops and asks whether you want it to run. Nothing is charged until you say yes.
Credits, honestly
- Same meter as the API. fetch spends the same credits as the REST API — same prices, same balance, one account.
- Every response reports the cost. Each fetch returns credits charged and credits remaining, so the assistant can tell you where you stand without you asking.
- Visible in your dashboard. MCP calls show up in usage and logs alongside API calls, tagged as MCP.
- Running out is explicit. A short balance returns an error naming the shortfall, with a link to top up — it never partially runs a job.
- Rate limits follow your credit pack, exactly as they do on the API.
- Failed calls are never charged.
Want to price a job yourself first? Use the calculator.
Sandbox responses
Some catalog endpoints do not yet have a connected data source. Those return clearly-labelled sample data and cost nothing. The response begins with a warning telling the assistant that the payload is sample data and must not be presented as real.
That warning is a safety feature, not a footnote. An assistant that quietly summarised invented follower counts as fact would be far worse than an error, so we make the shape of the response available for building against while making it impossible to mistake for live data.
Troubleshooting
New tools don't appear
The most common issue by far — clients cache the tool list. In Claude Code, run /mcp and choose Reconnect. In Claude web or desktop, disconnect the connector and add it again. Nothing is lost by reconnecting.
Asked to sign in again
The connection refreshes itself in the background. If it fully expires, reconnect once and you're done.
"Insufficient credits"
Your balance is below the cost of the job. Top up and rerun the prompt.
Wrong or empty results
Check the target format for that specific service with list_services. A username and a post URL are not interchangeable, and some services want an ID rather than a handle.
The client says it can't reach the server
Check the URL is exactly https://fastfetchz.com/mcp — no trailing slash, no /v1, and https rather than http. Clients that only run local servers need the mcp-remote bridge shown in the Windsurf, Cline & others tab.
Pack prices and rate limits live on the pricing calculator.