nohumans.directory — MCP Server

Find and vet paid x402 API services before an agent spends money on them.

A discovery and trust layer for paid x402 services. We verify listings two ways: by paying them real USDC and recording whether they deliver, and by continuously re-probing every listing's live 402 challenge — several times an hour, with the full history kept. Each listing carries a recency-weighted reliability score built from that history, plus on-chain payment evidence from our own index of the chain — so an agent can check whether an endpoint actually delivers before paying for it.

Free. No API key. No signup. No account.

Live stats: https://nohumans.directory/stats · REST API: https://api.nohumans.directory

---

Why this exists

Most x402 directories are catalogs: they list what exists. Some probe whether a 402 comes back, which proves an endpoint is alive — not whether anyone actually pays it, or whether it delivers when paid.

This one buys from listings and publishes what came back, per call, with the transaction hash. It probes every endpoint continuously and keeps the history: a listing moves unverified → verified only after sustained clean probes, and falls to failing when it stops responding. The reliability score is recency-weighted, so a service that broke yesterday can't hide behind a good month. And every listing shows its on-chain payment activity — payments, distinct paying wallets, dollars, per day — from our own index of USDC transfers on Base.

Every listing also carries a hand-written description of what the service actually does — which sounds mundane until you try semantic search against a catalog of bare URLs.

Installation

Remote server, streamable HTTP. No install, no build step, no credentials.

Claude Code

claude mcp add --transport http nohumans https://api.nohumans.directory/mcp

Claude Desktop / any stdio-only client

{
  "mcpServers": {
    "nohumans": {
      "command": "npx",
      "args": ["mcp-remote", "https://api.nohumans.directory/mcp"]
    }
  }
}

Direct remote connection

https://api.nohumans.directory/mcp

Transport: streamable-http · Authentication: none · Cost: free

Agent Skill

For agents that support the Agent Skills standard (Claude Code, Codex, and others), install the skill so the agent knows how to use the tools below intelligently — when to prefer verified listings, how to read score/status/x402_version/on-chain payment evidence, and how to report a real outcome via tx_hash:

mkdir -p ~/.claude/skills/nohumans-directory
curl -o ~/.claude/skills/nohumans-directory/SKILL.md \
  https://raw.githubusercontent.com/jalcodev/nohumans-mcp/main/skills/nohumans-directory/SKILL.md

Tools

find_paid_service

Search paid x402 services by capability, price ceiling, and minimum reliability score. Returns ranked results — verified services first, then by reliability, then by price.

Parameter	Type	Description
query	string	Free-text search across service names and descriptions
category	string	Exact category filter, e.g. data.fx, infra.trust
max_price	number	Maximum price per call, in USDC
min_score	number	Minimum reliability score, 0-1
limit	number	Max results, 1-50 (default 10)

get_service_details

Full record for a single service: pricing, accepted chains, request and response schemas, current status, reliability score, median latency, probe count, protocol version (x402_version), and on-chain payment evidence — 30-day totals, a per-day series (onchain_daily), how many listings share the payment address, and the sampled payment shape (x402-signed vs plain transfers vs other) — everything an agent needs before making a paid call.

Parameter	Type	Description
id	string	Listing ID, as returned by find_paid_service

resolve_endpoint

The reverse lookup: you already hold an endpoint URL — from another catalog, a prompt, a config file — and want this directory's verdict on it before paying. Returns the same full record as get_service_details when the URL is listed, a delisted warning if it was listed and later removed, or an explicit not-listed result. Not-listed means never checked here, not unsafe.

Parameter	Type	Description
url	string	The endpoint URL exactly as held, query string included

what_agents_are_asking_for

For sellers: the aggregate demand feed — what buyers searched this directory for, ranked by how many distinct days a term was asked (not by volume, which is trivially faked). Exclusions are published rather than hidden: queries under 5 characters, searches from clients that also edit listings here, and a floor of 3 distinct non-seller clients before a term appears. The window ends a day behind live and the response says when. Same data as GET /v1/demand.

No parameters.

Example

An agent that needs FX rates and won't pay more than half a cent, and refuses to touch anything with a reliability score under 0.9: