Connect an AI assistant (MCP)

You can plug Weird Requests straight into Claude, ChatGPT, or Gemini. Once connected, your assistant can browse the weirdest requests on the internet and answer questions about them in real time. No copy-pasting, no guessing.

The connection URL

https://weirdrequests.com/webmcp/v1

That's the whole secret. It's a public, read-only MCP server. No account, no API key. Your assistant can search and read; it cannot post, pay, or act on your behalf.

There's also a discovery descriptor at https://weirdrequests.com/.well-known/mcp.json that MCP-aware clients can read automatically.

What is MCP, in plain words

MCP (Model Context Protocol) is an open standard for letting AI assistants connect to outside services safely. Think of it as giving your assistant a library card for our site: it can look things up, but it can't make a mess.

What your assistant can do once connected

• Browse the feed. Trending, newest, or highest-paying open requests.
• Read a request. Its full description, price, deadline, and how many people have applied.
• Search. By keyword, category, or budget.

So you can just ask: "What are the weirdest open requests on Weird Requests right now?" and get a live answer.

How to connect

Claude

1. Open Settings → Connectors.
2. Choose Add custom connector.
3. Paste https://weirdrequests.com/webmcp/v1 and save.
4. Ask Claude about weird requests. It'll use the connection.

ChatGPT

1. Open Settings → Connectors (sometimes under Apps).
2. Add a new connector and paste the URL above.
3. Start a chat and ask it to browse Weird Requests.

Gemini

1. Open Gemini's extensions / connectors settings.
2. Add a custom MCP server using the URL above.
3. Ask Gemini to find you something weird.

Menu names shift as these apps update. If you don't see "Connectors," look for "MCP," "Apps," or "Extensions". The URL is the same everywhere.

For developers

The endpoint speaks the standard MCP transport over HTTP. The public tool catalog (and JSON shapes) is advertised in the /.well-known/mcp.json descriptor. The public surface is rate-limited per IP and requires no auth; authenticated, write-capable tooling is a separate OAuth-scoped surface.

There's also an /llms.txt index (with a key-facts block covering fees, limits, and timers), an /llms-full.txt with the complete documentation inlined in one fetch, and a raw-markdown version of every doc page (append .md to any docs URL) — so agents can read the documentation itself, not just the data.