Pinecone ships a Developer MCP server aimed squarely at the loop of building a RAG application: an agent can read current Pinecone documentation, create an index with integrated embedding, upsert records, run semantic search and rerank the results — all from the same conversation where it is writing the code.
Bring your own
Claude Sonnet 4.5
Chat with 60+ AI models on the same workflow — switch to a different model mid-conversation and re-run the same prompt, or use Compare mode to put several side-by-side and balance quality vs. cost.
A Pinecone API key from the console. Documentation search works without one; everything touching an index needs it.
How models use it and what it is built for.
Two things make this server useful rather than merely convenient. First, search-docs puts current Pinecone documentation in front of the model, which matters because the SDK surface has changed substantially across versions and a model working from memory will write code against an API that no longer exists. Second, create-index-for-model creates an index with integrated inference configured, meaning Pinecone embeds text server-side — so upsert-records and search-records take plain text and the agent never constructs a vector. On top of that, cascading-search queries several indexes and merges the results, and rerank-documents applies a reranking model to a result set, which is the cheapest meaningful quality win in most retrieval pipelines. Pinecone also publishes a separate Assistant MCP server for querying a Pinecone Assistant knowledge base; the Developer server is the one you want when you are building.
Typical tools an AI model can call. Exact names vary by version.
Taken from the official Pinecone documentation — see Pinecone MCP server — official documentation for the full reference.
Environment variables
PINECONE_API_KEYrequiredAPI key from the Pinecone console. Scope it to the project holding the indexes the agent should reach.
Client configuration
npx (recommended)
The Developer MCP server — docs, indexes, records and reranking.
{
"mcpServers": {
"pinecone": {
"command": "npx",
"args": ["-y", "@pinecone-database/mcp"],
"env": { "PINECONE_API_KEY": "YOUR_API_KEY" }
}
}
}Copy any of these into MCP Agent Studio after connecting.
Create an index for this dataset and upsert these documents into it.
Search the docs for how namespaces interact with metadata filtering, then show me the code.
Query the index for passages about refund policy and rerank the top 20.
How many records are in each namespace of my production index?
This is not a single-model product: you get the same MCP connection with 60+ models (Claude, GPT, Gemini, DeepSeek, open-weight, and more), you can switch mid-conversation, and you can open Compare mode to run the same prompt against multiple models at once. The card above is a suggested starting point for this server — not the only choice.
Default pick for Pinecone
Claude Sonnet 4.5
This server is used while writing retrieval code. Sonnet 4.5 checks the docs tool before generating SDK calls, which is exactly the behaviour that makes it worth installing.
Listing tools proves the server is reachable, not that a model can work with it. Evals go further: they read every tool on the server, write a test suite from its real schemas, and run it — code decides pass/fail on the responses (schema conformance, error codes, pagination, result caps) while a scoring model grades plain-English tasks driven through the tools.
Get a pass/fail report per tool with the evidence behind each verdict — and replay the same suite after every schema change. Destructive tools are excluded from the run.
Run evalsOpen MCP Agent Studio with the connection pre-filled. Add your token, pick any of 60+ models, and start chatting — no install required.
Open Agent StudioCommon questions about connecting, scoping and using it safely.
It is Pinecone’s official Developer MCP server, published as @pinecone-database/mcp. It gives an AI coding agent tools to search Pinecone documentation, create and inspect indexes, upsert records, run semantic search and rerank results.
Not if the index was created with create-index-for-model. That configures integrated inference, so Pinecone embeds text server-side and upsert-records and search-records both work in plain text. For an index you created yourself without integrated embedding, you still supply vectors.
The Developer server is for building — docs, index management, records, search. The Assistant server queries a Pinecone Assistant knowledge base you have already built. Install the Developer server while writing the application, the Assistant one when consuming it.
Because the Pinecone SDK has changed shape across major versions, and a model relying on training data will confidently produce calls from an older API. Searching the live documentation first is what stops generated retrieval code from failing on the first run.
Pinecone is fully managed with integrated embedding and reranking, so there is less to operate. Qdrant can run embedded on local disk with no service at all and is open source. For a hosted production RAG stack Pinecone is the lower-effort choice; for local-first or self-hosted work, Qdrant.