For developers · MCP

Give your agent First Instinct as a native tool

First Instinct ships a Model Context Protocol server. Add it to any MCP client, Claude, Cursor, or your own agent, and the model can resolve upstream evidence and evaluate agent identities directly, offline and with no API key. It is the fastest way to make “should this agent action be accepted?” a question your agent can weigh mid-task.

Install

Add First Instinct to your MCP client config (this example is Claude Desktop's claude_desktop_config.json), then restart the client.

{
  "mcpServers": {
    "idn": {
      "command": "npx",
      "args": ["-y", "@idn-network/mcp"],
      "env": { "IDN_REGISTRY_URL": "https://inct.world" }
    }
  }
}

The server runs over stdio via npx, so there is nothing to host. It talks to the public evidence resolver at inct.world; point IDN_REGISTRY_URL elsewhere for a private deployment.

Tools it exposes

  • idn_resolve(handle)

    Resolve a handle to its latest signed evidence: issuer, status, expiry, and the attestation JWS.

  • idn_verify_agent(handle, minKycTier?, minStrength?, minTenure?)

    Resolve, check the signature offline against the issuer's key, check freshness, and apply an optional policy. Returns a reproducible evaluation the agent can weigh before it acts.

  • idn_list_issuers(none)

    List the regulated evidence sources whose signatures back attestations, and their status.

Evaluation is offline: the server fetches the issuer's public key and checks the attestation's Ed25519 signature itself, so a positive result is cryptographic, not a trusted API response. For the raw HTTP contract, read the agent skill or the docs.