Skip to content

What is Agent Finder?

Agent Finder is an open standard for discovering agent extensibility mechanisms. It enables anyone to build a search engine for skills, MCPs, subagents, and A2A capabilities, making them discoverable to any agent harness—whether it be Gemini, Claude Code, Copilot, Codex, or Manus.

Get Started Read the Specification

Why Agent Finder?

Discovering capabilities to enhance your agent (skills, subagents, hooks) is currently a manual, developer-locked process. With the volume of tools and agents exploding into the millions, sifting through them or pre-installing thousands of integrations is no longer practical. Even with basic search, traditional tool selection degrades in accuracy and latency when context windows are stuffed with thousands of schemas.

What Agents Should Do

  1. Global Access: Access the world's growing catalog of tools and capabilities.
  2. Zero Prompt Bloat: Maintain minimal active toolsets to prevent context window degradation.
  3. Billions-Scale Discovery: Find the exact tool for the right task out of a catalog of billions.
  4. Decentralized Trust: Establish cryptographic trust relationships with authors and sources.
  5. On-the-Fly Execution: Install and invoke these capabilities dynamically at runtime.

What Agent Finder Enables

  • 🔌 Federated Search: Stand up a decentralized search engine for any agent capability (MCP, skills, A2A, and custom APIs).
  • 🌐 Self-Sovereign Publishing: Host your standard ai-catalog.json at a well-known domain root (/.well-known/) to advertise your capabilities.
  • ⚡ Cross-Harness Portability: Enable any agent from any harness to dynamically discover, verify, and use your tools.

Quick Preview

Here is how a publisher advertises an MCP server using the standard ai-catalog.json manifest:

{
  "specVersion": "1.0",
  "host": {
    "displayName": "Acme AI",
    "identifier": "did:web:acme.com"
  },
  "entries": [
    {
      "identifier": "urn:ai:acme.com:agent:trading",
      "displayName": "Finance Trading Agent",
      "type": "application/mcp-server+json",
      "url": "https://api.acme.com/agent.json",
      "representativeQueries": [
        "analyze market trend for 2026"
      ]
    }
  ]
}