proofmarket-mcp

An MCP server that exposes a machine-callable proof-market surface. An agent connects over the Model Context Protocol, submits Proof Requests, reads the market's supply / allocated / utilized, and composes multi-step workflows that run with no human in the loop.

It is the surface Fermah keeps describing: the primary user of a protocol is no longer a human clicking a dashboard, it is an agent calling an execution surface directly. This repo is a small, runnable answer to "what does that surface look like."

This is a reference model, not Fermah's production Kernel and not live network data. The coordination mechanics come from proofmarket-sim; the same engine settles every number. Vocabulary follows Fermah: Seekers submit Proof Requests, the Matchmaker assigns them to Prover Nodes.

Tools

• market_state - snapshot supply, allocated, utilized, and the matching and scheduling gaps of the running market.

• simulate_coordination - run the coordination model over a full horizon for a regime (today / no-operator / coordinated). Same supply in every regime; only coordination changes.

• submit_proof_request - a Seeker submits a Proof Request; returns an id to poll.

• poll_proof - check a request; on settle it carries a proof id and attestation, delivered the way a proof returns to a Seeker callback. Zero humans.

• compose_workflow - define an ordered sequence (observe, request_proof, act, settle) and the surface runs the whole thing, returning an attested trace with humans in the loop = 0.

There is also an proofmarket://about resource describing what the surface is and is not.

Run it

npm install
npm test          # in-memory MCP client lists tools, drives the surface
npm run demo      # an agent drives the surface end to end, no human in the loop
npm start         # start the MCP server on stdio

Connect it to an agent

Claude Desktop (claude_desktop_config.json), Cursor, or any MCP client over stdio:

{
  "mcpServers": {
    "proofmarket-surface": {
      "command": "node",
      "args": ["/absolute/path/to/proofmarket-mcp/src/server.js"]
    }
  }
}

Then ask the agent to check the market, compare coordination regimes, and compose a proof workflow. It will call the tools directly, with no human in the loop.

Why this exists

A protocol without a machine-callable surface is invisible to agents no matter how capable its onchain logic is. The point of this repo is not to reimplement Fermah. It is to make the surface concrete, runnable, and honest: an agent can drive a proof market end to end, and you can read every line of how it works.

License

MIT. Built by ZERTH MAXX (@0maxxdev). Not affiliated with or endorsed by Fermah.