Pinecone MCP Server: Setup Guide, Use Cases & Limitations

What Is Pinecone MCP Server?

Pinecone MCP Server connects Claude Code to Pinecone vector databases, enabling you to query indexes, manage embeddings, and build retrieval-augmented generation (RAG) pipelines directly from your Claude sessions. It is maintained by sirmews as an open-source project on GitHub.

Quick Facts

What You Can Do With It

• Vector search: Query a Pinecone index by vector to retrieve semantically relevant results from your stored data.
• RAG pipelines: Feed retrieved context directly into Claude prompts, letting you build retrieval-augmented generation workflows without leaving your coding session.
• Semantic similarity: Compare embeddings across your index to find conceptually related records, useful for deduplication or recommendation tasks.
• Embedding management: Upsert, update, and delete vectors in your Pinecone indexes, keeping your data layer in sync as your corpus changes.

How to Set It Up

Prerequisites: You need a Pinecone account with at least one index created and an active API key.

1. Verify Node.js is installed (v18+ recommended):

node --version
**2. Add the server to your Claude Code configuration.**

Open or create `.claude/mcp.json` in your project root (or `claude_desktop_config.json` for the desktop app) and add the following:

```json
{
  "mcpServers": {
    "pinecone": {
      "command": "npx",
      "args": ["mcp-pinecone"],
      "env": {
        "PINECONE_API_KEY": "your-api-key-here",
        "PINECONE_INDEX_NAME": "your-index-name"
      }
    }
  }
}

Replace your-api-key-here with your Pinecone API key and your-index-name with the target index.

3. Restart Claude Code to load the new server configuration. You should see the Pinecone server listed in the MCP tools panel.

4. Test the connection by asking Claude to query your index — for example: “Search my Pinecone index for documents related to machine learning.”

Works Best With

Pinecone MCP Server is confirmed compatible with Claude Code. If you use Claude Code as your primary interface, this server integrates directly into your existing MCP tool configuration without additional setup. There is no documented compatibility with other MCP clients at this time — check the GitHub repo for updates as the ecosystem evolves.

Known Limitations

• Authentication is required: The server will not start without a valid PINECONE_API_KEY. Ensure the key has read/write permissions for the target index.
• Single-index scope per config: The PINECONE_INDEX_NAME environment variable targets one index at a time. Working across multiple indexes may require separate server entries in your config.
• Rate limits are Pinecone-side: Query and upsert rate limits are governed by your Pinecone plan tier, not the MCP server itself. High-frequency RAG workflows on free-tier plans may hit throttling.
• Embedding generation is external: This server manages vectors already in Pinecone; it does not generate embeddings. You need a separate embedding pipeline (OpenAI, Cohere, etc.) to produce vectors before upserting.
• No offline mode: All operations require a live connection to Pinecone’s API.

Community Resources

• GitHub Repository — primary source for documentation, changelog, and setup details
• GitHub Issues — report bugs or check for known problems before filing a new issue
• Pinecone Community Forum — general Pinecone questions, including integration patterns relevant to MCP workflows