Pinecone MCP server
Pinecone's official developer MCP server: search indexes, manage records, rerank results, and look up Pinecone docs from your agent.
The Pinecone MCP server is Pinecone's official developer integration that lets an AI agent work directly against your Pinecone vector database and documentation. The agent can list and describe indexes, read index statistics and namespaces, create a new index backed by an integrated embedding model, upsert and search records using that integrated inference (so you pass text and Pinecone handles the embeddings), and improve relevance with reranking. It can also run a cascading search across multiple indexes with deduplication and reranking, and search the official Pinecone documentation to ground setup and API questions, making it a strong backend for retrieval-augmented agents and a co-pilot for building on Pinecone itself.
The developer server is published on npm as @pinecone-database/mcp and runs locally over stdio via npx (npx -y @pinecone-database/mcp). It authenticates with a PINECONE_API_KEY from the Pinecone console and talks to api.pinecone.io for data operations and docs.pinecone.io for documentation search. Separately, Pinecone offers an Assistant MCP server: every Pinecone Assistant has a fully managed remote MCP endpoint (https://<assistant-host>/mcp/assistants/<name>) authenticated with a bearer API key and exposing a context retrieval tool, with a self-hostable Docker image also available. Choose the developer server to operate indexes and records directly, or the Assistant endpoint for managed context retrieval over uploaded files.
Quick install
Copy-paste configs are provided for all 8 supported clients. Pick your client below.
Add to ~/.claude.json
{
"mcpServers": {
"pinecone": {
"command": "npx",
"args": [
"-y",
"@pinecone-database/mcp"
],
"env": {
"PINECONE_API_KEY": "<PINECONE_API_KEY>"
}
}
}
}claude mcp add pinecone -- npx -y @pinecone-database/mcpAvailable tools
| Tool | Description |
|---|---|
| search-docs | Searches the official Pinecone documentation. |
| list-indexes | Lists all Pinecone indexes in the project. |
| describe-index | Describes the configuration of a specific index. |
| describe-index-stats | Reports statistics about an index, including record count and available namespaces. |
| create-index-for-model | Creates a new index that uses an integrated inference model to embed text as vectors. |
| upsert-records | Inserts or updates records in an index using integrated inference. |
| search-records | Searches records in an index from a text query using integrated inference, with metadata filtering and reranking options. |
| cascading-search | Searches records across multiple indexes, deduplicating and reranking the combined results. |
| rerank-documents | Reranks a set of records or text documents using a specialized reranking model. |
Required configuration
- PINECONE_API_KEYRequired
Pinecone API key from the Pinecone console, used by the developer MCP server. Required.
What you can do with it
Build a retrieval-augmented agent on Pinecone
Create an index with create-index-for-model, upsert your content with upsert-records, then let the agent answer questions with search-records and rerank-documents, all using Pinecone's integrated embeddings so no separate embedding step is needed.
Operate and debug Pinecone from your editor
List and describe indexes, read describe-index-stats to see record counts and namespaces, and call search-docs to look up the exact API or configuration option, without leaving the agent to open the console.
FAQ
- Is it free?
- The MCP server is free and open source, but it operates against Pinecone, which has a free starter tier and paid plans. Index storage, queries, integrated embedding, and reranking are billed against your Pinecone account and PINECONE_API_KEY.
- Does it support remote/OAuth?
- The developer server (@pinecone-database/mcp) runs locally over stdio with an API key. Separately, Pinecone Assistant exposes a fully managed remote MCP endpoint authenticated with a bearer API key (no OAuth), which you can also self-host via Docker.