fetcher.cloud — the web capability layer for AI agents

One MCP server and API that gives an autonomous AI agent the public web. Render and screenshot pages, extract clean content, resolve links, parse feeds, and inspect domains, DNS, SSL, and WHOIS — 18 capabilities, paid per-call in USDC via x402, with a free daily tier.

Whenever an agent needs something from the internet, it uses fetcher.

Why it exists

LLMs are brilliant at reasoning and blind at retrieval. The moment real-world data needs a browser, a payment, or a login, an agent stalls. fetcher is the specialist capability layer it calls — a single interface for everything an agent can't do alone, billed the only way that works for software: per call, in stablecoins, with no account or API key.

Related MCP server: nanoparse-mcp

How agents use it

1. Discover — GET /capabilities returns every capability with category, pricing, examples, and schemas. Or connect the MCP server at /mcp.

2. Call — a normal GET request. The first unpaid call past the free tier returns 402 Payment Required with machine-readable terms.

3. Pay — the agent's wallet pays a fraction of a cent in USDC on Base and retries. No human in the loop.

Works with Conway Automaton, Claude, OpenAI Agents, Google ADK, any MCP client, or any agent that can make an HTTP request.

The capabilities

Full catalog with pricing and examples: https://api.fetcher.cloud/capabilities

Machine-readable surfaces

/ (manifest) · /capabilities · /pricing.json · /openapi.json · /mcp · /.well-known/ai-plugin.json · llms.txt

Repository layout

worker/         # the Cloudflare Worker (API, payments, MCP)   → see worker/README.md
website/        # the landing page (Cloudflare Pages)
test-client/    # a script that pays an endpoint on Base mainnet
conway-skill/   # ready-to-submit Conway Automaton skill

Deploy

See worker/README.md and DEPLOY.md for step-by-step deployment (Cloudflare Workers + Pages). Payments require Coinbase CDP facilitator keys set as Worker secrets.

License

MIT — see LICENSE.