Issue #6 of the Agent Native Registry newsletter. If someone forwarded this to you, subscribe at https://agentnativeregistry.com

This week I scored every major database and vector database for AI agent compatibility. The results were not what I expected.

The surprise: the databases marketed as "AI-native" (Pinecone, Weaviate) score LOWER than traditional databases (Turso, Neon) on our Agent Native scoring rubric.

THE LEADERBOARD: Databases & Vector DBs

Turso: 87/100 (MCP: yes) - SQLite at the edge, agent-optimized Neon: 85/100 (MCP: yes) - Serverless Postgres, best free tier Qdrant: 83/100 (MCP: yes) - Best vector DB for agents Supabase: 82/100 (MCP: yes) - Only DB with MCP + llms.txt Pinecone: 78/100 (MCP: no) - Most popular vector DB, no MCP server Weaviate: 72/100 (MCP: no) - Lowest agent tooling score (65/100)

THE PARADOX

Pinecone and Weaviate are purpose-built for the AI era. They exist to store embeddings for LLM applications. Yet they score lower on agent-native compatibility than Turso, a SQLite edge database that predates the AI wave.

Why? Because "built for AI" means they're built for humans building AI apps. They optimized for Python SDKs, hosted dashboards, and developer experience. That's great for a person spinning up a RAG pipeline. It's not great for an AI agent that needs to:

• Discover the tool programmatically
• Create an account without clicking OAuth buttons
• Call the API without managing a browser session

Turso has an MCP server. Qdrant has an MCP server. Pinecone and Weaviate don't. That single gap accounts for most of the score difference.

QDRANT: The agent-native vector DB

Qdrant scores 83/100, the highest of any dedicated vector database. Open source, official MCP server (github.com/qdrant/mcp-server-qdrant), generous free tier with programmatic account creation.

Agent tooling score: 88/100 -- higher than Supabase, Neon, and most other databases.

The gap: no llms.txt yet. Add that and Qdrant pushes toward 90+.

PINECONE: Popular but an agent trap

Pinecone scores 78/100. Its biggest penalty: pricing (70/100). The free tier limits you to a single index. Scaling requires upgrading before an agent can do meaningful work across multiple workloads.

No MCP server means agents have to write raw REST calls. With $100M+ raised, Pinecone has the resources to fix this. They've chosen not to yet.

SUPABASE: The sleeper pick

Supabase scores 82/100 and is the only database in our dataset with both an MCP server AND a llms.txt file.

For agents that need vector similarity search, pgvector (built into Supabase) gives you semantic search without switching databases. If your data is already in Postgres, there's a real argument for skipping Pinecone entirely and using Supabase pgvector.

WHAT THIS MEANS FOR AGENT BUILDERS

Use Qdrant for a dedicated vector database. Use Supabase pgvector if you're already on Postgres. Skip Weaviate unless you have a specific reason.

The broader pattern: the tools with the best agent-native scores aren't always the famous ones. They're the ones that invested in MCP servers, clean API keys, and discoverable documentation.

Full data: https://agentnativeregistry.com/best-database-tools-for-agents.html All 100 tools: https://agentnativeregistry.com

Want to see your tool rated? Reply to this email with your tool's URL.