Creativity · MCP — server

MCP Pinecone Server

The MCP Pinecone Server connects Model Context Protocol clients to a Pinecone vector database. With an API key scoped to a specific index, Claude Desktop or Cursor can search for top-k nearest neighbours, upsert embeddings, and pull document metadata — turning any private corpus into an LLM-searchable knowledge base over stdio.

MCP facts

Kind: server
Ecosystem: anthropic-mcp
Language: TypeScript / Node.js
Transports: stdio

Capabilities

Tools: search_records — top-k semantic search with optional metadata filter
Tools: upsert_records for adding documents, delete_records for cleanup
Tools: describe_index, list_indexes for index introspection
Works with Pinecone serverless and pod-based indexes

Install

npx -y @pinecone-database/mcp

Configuration

{
  "mcpServers": {
    "pinecone": {
      "command": "npx",
      "args": ["-y", "@pinecone-database/mcp"],
      "env": {
        "PINECONE_API_KEY": "pcsk_xxx",
        "PINECONE_INDEX_NAME": "docs-prod"
      }
    }
  }
}

Frequently asked questions

Do I need to run my own embedding pipeline?

Yes — the MCP server is for query and upsert. Use a batch ETL (e.g. a Python script or a Vercel/Cloudflare Worker) to chunk, embed, and push documents. The agent then searches the index on demand.

Which embedding model should I pair it with?

Pinecone is model-agnostic. Common pairings are OpenAI text-embedding-3-large, Cohere embed-v3, and Voyage voyage-3 — pick one, match its dimension to the index, and keep it consistent for upserts and queries.

Can I scope the API key to a single index?

Yes — in Pinecone's console create a project-scoped key or, for fine-grained control, a project with only the target index. Avoid org-level keys in agent configs.

Sources

Pinecone MCP server — accessed 2026-04-20
Pinecone API reference — accessed 2026-04-20