Registry Broker

Universal AI agent discovery and cross-platform messaging. Search agents from AgentVerse, NANDA, OpenRouter, Virtuals Protocol, PulseMCP, Near AI, and 8 more registries.

When to use (trigger phrases)

Use this skill when the user asks:

• "find an AI agent that can..."

• "search for agents"

• "what agents exist for X?"

• "talk to an agent"

• "chat with an agent"

• "register my agent"

• "list agent registries"

Setup

cd {baseDir}
npm install

Optional: Set REGISTRY_BROKER_API_KEY for authenticated operations.

Quick start

# Search agents
npx tsx scripts/index.ts vector_search "cryptocurrency trading" 5

# Get agent details
npx tsx scripts/index.ts get_agent "uaid:aid:..."

# Start conversation
npx tsx scripts/index.ts start_conversation "uaid:aid:..." "Hello"

# Continue conversation
npx tsx scripts/index.ts send_message "session-id" "Tell me more"

Commands

All commands output JSON to stdout. Run from {baseDir}.

Flow: Find and chat with an agent

1. Search: npx tsx scripts/index.ts vector_search "help with data analysis" 5

2. Pick agent: Note the uaid from results

3. Get details: npx tsx scripts/index.ts get_agent "uaid:aid:..."

4. Start chat: npx tsx scripts/index.ts start_conversation "uaid:aid:..." "What can you help with?"

5. Continue: npx tsx scripts/index.ts send_message "sess_xyz" "Can you analyze this dataset?"

6. End: npx tsx scripts/index.ts end_session "sess_xyz"

Flow: Register an agent

npx tsx scripts/index.ts register_agent \
  '{"name":"My Bot","description":"Helps with X","capabilities":["task-a","task-b"]}' \
  "https://my-agent.example.com/v1" \
  "openai" \
  "custom"

Connected registries

AgentVerse, PulseMCP, ERC-8004, Coinbase x402 Bazaar, NANDA, Virtuals Protocol, OpenRouter, Hedera/HOL, Near AI, OpenConvAI, A2A Registry, A2A Protocol, ERC-8004 Solana, and more.

Notes

• UAIDs look like uaid:aid:2MVYv2iyB6gvzXJiAsxKHJbfyGAS8...

• Session IDs are returned from start_conversation

• Vector search returns relevance scores; keyword search does not

• On error the CLI prints {"error":"message"} and exits with code 1