ClawMarket Agent Matcher

Tell us your business problem. Our Gemini + MongoDB agent searches ClawMarket's live AI agent catalog and recommends the right agent, what it costs, and links you straight to purchase.

Comment

Inspiration

ClawMarket is a live marketplace where AI agents offer real services for real money. USDC has settled on Base. Buyers exist. But discovery was still manual. You had to browse to find the right agent for your problem.

We wanted to fix that. Tell the agent your problem in plain English, and it finds the right match in the catalog.

What it does

ClawMarket Agent Matcher is a Gemini-powered agent that takes any business problem or task description and searches the live ClawMarket catalog to recommend the best agent and skill. Every recommendation includes what it costs in USDC or MARKS and a direct link to purchase.

The demo is not a mock. The catalog is live. The purchase links are real.

How we built it

The core is a Gemini agent built with the Google Agent Development Kit (ADK), connected to MongoDB Atlas through the MongoDB MCP server. The agent searches the seeded catalog collection, reasons about fit, and hydrates prices and purchase links from live data so nothing is model-invented.

The matching pipeline has three tiers with graceful degradation: the ADK agent with MongoDB MCP as the primary path, MongoDB Atlas vector search with a Gemini rerank as the second tier, and keyword relevance against the live ClawMarket API as the final fallback. Embeddings use gemini-embedding-001 at 768 dimensions. The frontend is Next.js deployed on Vercel.

Challenges we ran into

Getting the agent to consistently surface the right match across 30+ skills in 15 categories required careful prompt tuning. The catalog is diverse, ranging from paranormal to corporate intelligence, so generic retrieval was not enough.

Free-tier Gemini rate limits forced us to design the fallback chain early. Migrating embedding models mid-build meant re-seeding the vector index and rewriting the batching logic, since gemini-embedding-001 embeds per document rather than in batches.

Accomplishments that we're proud of

Every result a judge clicks leads to a real, purchasable service on a marketplace where real USDC has settled. Grounding the agent in live database reads instead of model recall eliminated hallucinated prices and links entirely.

What we learned

MongoDB MCP integration is powerful but the data shape matters a lot. Clean, structured fields in the collection make the agent dramatically more accurate. And an agent that degrades gracefully beats one that is impressive only when everything works.

What's next for ClawMarket Agent Matcher

Natural language filtering by price, category, and trust tier. And agent-to-agent hiring: ClawMarket already supports agents purchasing from other agents over its API, so the Matcher's next step is to not just recommend but autonomously hire on a buyer's behalf.

Built With

  • base
  • clawmarket-api
  • gemini
  • google-cloud-agent-builder
  • mongodb-atlas
  • mongodb-mcp
  • next.js
  • vercel
Share this project:

Updates