Set up Pinecone MCP Server

Pinecone MCP Server connects coding assistants like Cursor, Claude Code, and Claude Desktop directly to Pinecone projects and their official documentation. According to official Pinecone documentation, a connected agent can search documentation, manage indexes, upsert records, and run semantic searches without switching between the assistant and the Pinecone console. This guide shows how to set up the server safely, what prerequisites are needed, and what to watch for when using it in development.

Check prerequisites

Before setup, three things must be in place. First, Node.js v20 or later must be installed. Verify this with node --version. Second, npx must be available in PATH; check with which npx. Third, you need a Pinecone API key, which can be created through the Pinecone console. Without an API key, only documentation search is available according to the documentation; index operations are not possible. Have the key ready but do not store it in configuration files until the storage location is secured.

Also check which client you are using. Cursor, Claude Desktop, Claude Code, and Antigravity are explicitly described in the official documentation. For Gemini CLI, the repository documents an extension install command. Verify that your client has current MCP support before starting configuration.

Store the API key securely

The PINECONE_API_KEY belongs in the client's secret or credential management facility, not in source code, not in a shared configuration file, and not as plaintext in terminal history. Most clients provide an environment-variable management or secret function that can make the key available to the MCP process. Use that instead of inserting the key directly into JSON configuration files that could accidentally be checked into a repository.

If a key has already been placed in a configuration file, add the file to .gitignore immediately before it reaches a repository. Rotate or revoke a potentially exposed key through the Pinecone console. The key grants read and write access to your Pinecone project; an exposed key should be revoked without delay.

Set up Cursor

For a project-level setup, create a .cursor/mcp.json file in the project root and add the following configuration:

{
  "mcpServers": {
    "pinecone": {
      "command": "npx",
      "args": ["-y", "@pinecone-database/mcp"],
      "env": {
        "PINECONE_API_KEY": "<your-api-key>"
      }
    }
  }
}

Replace <your-api-key> with the actual key, or use a Cursor-specific secret facility. For a machine-wide setup, place the file in the home directory at .cursor/mcp.json. Check the server status under Cursor Settings > MCP. The official documentation also recommends adding Cursor rules for correct usage of the MCP server.

Set up Claude Desktop

In Claude Desktop, open Settings > Developer > Edit Config and insert the configuration into claude_desktop_config.json. The structure matches the Cursor example above. After saving, restart Claude Desktop. A hammer icon should appear on the new chat screen, showing the available MCP tools. If the icon is absent, check the JSON syntax of the configuration file and verify that Node.js is installed correctly.

Set up Claude Code

For Claude Code, use the following command:

claude mcp add-json pinecone-mcp \
  '{"type": "stdio",
    "command": "npx",
    "args": ["-y", "@pinecone-database/mcp"],
    "env": {"PINECONE_API_KEY": "YOUR_API_KEY"}}'

After restarting Claude Code, check the status with /mcp. The official documentation also describes the Pinecone Claude Code Plugin, which bundles the MCP server, eight skills, and slash commands, if a more comprehensive integration is desired.

Run a first test

After setup, start with a low-risk documentation query, for example: "Search the Pinecone docs for information about metadata filtering." This verifies that the assistant recognizes and uses the tool before running any write operations. If that works, use list-indexes to list your own indexes and verify project access.

For initial write operations, choose a test environment or a namespace that does not contain production data. Create an index with create-index-for-model, upsert test data, and then search it. This lets you evaluate queries and results directly in the development environment before real content is upserted.

Know the server's limits

According to official documentation, the Pinecone Developer MCP Server only supports indexes with integrated inference. External embedding models and plain vector search without integrated inference are not supported. For access to Pinecone Assistants and their knowledge bases, there is a separate Pinecone Assistant MCP Server that is configured independently. Keep these two use cases clearly separate to avoid configuration errors.

Write permissions and human approval

Operations such as upsert-records and create-index-for-model permanently modify the Pinecone project. In a production environment, explicit approval from a responsible person should be part of the process before the assistant executes such changes. Restrict agent permissions to the operations that are actually needed. Data upserted through upsert-records leaves the local device and is stored on Pinecone servers; review Pinecone's privacy policy and terms of service directly with the provider before using sensitive content. This guide intentionally gives no price amounts.

FAQ

Which npm package starts the Pinecone MCP Server? @pinecone-database/mcp, launched via npx -y @pinecone-database/mcp.

Do I need an API key for documentation search? No. Documentation search is available without an API key according to official documentation. A key is required for index operations.

How do I check whether the server is active? In Cursor under Settings > MCP; in Claude Code with the /mcp command.