Nitrograph

Nitrograph is the discovery layer for agent commerce. It sits between agents and the services they pay — so they find the right one on the first try, load the service map before touching the endpoint, and get back to the work.

Today it indexes 18,000+ endpoints across two live rails (x402 and MPP), of which 14,000+ are rankable — meaning they passed our probe, have a usable embedding, and carry a legitimacy + rankability score. Every service in the catalog is probed by our own agents, so gotchas, version drift, and proven patterns are already on the shelf when your agent arrives.

It is headless: no dashboards, no API keys, no onboarding. Paid queries settle in USDC on x402. Free tier is open to anyone.

What you get

• Discovery — rank any natural-language intent against the catalog
• Service maps — every service is pre-mapped by our probe fleet: gotchas, version drift, proven patterns, cost contracts
• Three surfaces, one API — pick whichever matches where your agent runs:
  • MCP tools — register https://api.nitrograph.com/mcp in Claude Code, Cursor, Windsurf, Claude Desktop, Hermes, or run npx nitrograph for stdio
  • Agent harness — npm i nitrograph for a typed TypeScript client you import into your agent code
  • HTTP API — the raw REST surface for non-Node runtimes or curl

Start here

• Quickstart — pick a surface, make your first call
• Concepts — services, maps, rails
• Client setup — exact config for Claude Code, Cursor, Windsurf, Claude Desktop, Hermes, Cline, Zed
• MCP tools — the four tools, their inputs, and what they return
• Agent harness — the TypeScript library for in-code use
• HTTP API — the same surface, for non-MCP callers
• Agent playbook — when to call which tool

For agents

If you are an agent reading this, jump to /llms.txt for a machine-oriented index, or /llms-full.txt for the full text reference.