turbopuffer MCP server

OfficialturbopufferConfig last verified Jun 1, 2026

turbopuffer's official MCP server helps an agent build on turbopuffer: search the docs and run TypeScript SDK code against your namespaces in a sandbox.

The turbopuffer MCP server is turbopuffer's official integration for building on turbopuffer, the serverless vector and full-text search database built on object storage. Rather than wrapping each REST endpoint as its own tool, it gives an agent two high-leverage capabilities. A documentation search tool lets the agent query turbopuffer's API and SDK docs in natural language so it can ground answers in current guidance, and a code-execution tool lets the agent write TypeScript against the official turbopuffer SDK and run it in an isolated sandbox — so it can list namespaces, write and query documents, inspect schema, and tune recall by actually exercising the SDK rather than guessing at parameters. This design makes the server a strong copilot for integrating turbopuffer into an application and for operating namespaces from inside your editor.

The server is published on npm as @turbopuffer/turbopuffer-mcp and runs locally over stdio via npx (npx -y @turbopuffer/turbopuffer-mcp@latest), authenticating with a TURBOPUFFER_API_KEY and an optional TURBOPUFFER_REGION and default namespace. turbopuffer also hosts a remote Streamable HTTP endpoint at https://turbopuffer.stlmcp.com, where the API key is supplied via the x-turbopuffer-api-key header or a standard Authorization: Bearer token, so the same tools are reachable without a local install. The server is in beta and built on the Stainless MCP framework.

Quick install

Copy-paste configs are provided for all 8 supported clients. Pick your client below.

Add to ~/.claude.json

~/.claude.json
json
{
  "mcpServers": {
    "turbopuffer": {
      "command": "npx",
      "args": [
        "-y",
        "@turbopuffer/turbopuffer-mcp@latest"
      ],
      "env": {
        "TURBOPUFFER_API_KEY": "<TURBOPUFFER_API_KEY>",
        "TURBOPUFFER_REGION": "<TURBOPUFFER_REGION>",
        "TURBOPUFFER_DEFAULT_NAMESPACE": "<TURBOPUFFER_DEFAULT_NAMESPACE>"
      }
    }
  }
}
Or via CLI
bash
claude mcp add turbopuffer -- npx -y @turbopuffer/turbopuffer-mcp@latest

Available tools

ToolDescription
search_docsSearches turbopuffer's API and SDK documentation in natural language, returning relevant guidance at a chosen level of detail.
executeRuns agent-written TypeScript against the official turbopuffer SDK in an isolated sandbox, letting it list namespaces and write, query, and inspect documents directly.

Required configuration

  • TURBOPUFFER_API_KEYRequired

    turbopuffer API key used to authenticate SDK calls. Required.

  • TURBOPUFFER_REGIONOptional

    turbopuffer region for your data (e.g. gcp-us-central1). Optional.

  • TURBOPUFFER_DEFAULT_NAMESPACEOptional

    Default namespace the agent should operate against. Optional.

What you can do with it

Integrate turbopuffer into an app

Ask the agent how to model and query a namespace; it searches the docs for the right API, then writes and runs TypeScript SDK code in the sandbox to verify the approach against your real data before you ship it.

Operate namespaces from your editor

Use the execute tool to list namespaces, upsert documents, run vector and full-text queries, and check schema or recall, so you can manage and debug a turbopuffer index without switching to a separate console.

FAQ

Is it free?
The MCP server is free to use, but it operates against turbopuffer, which has its own pricing. Storage, writes, and queries are billed to your turbopuffer account via the TURBOPUFFER_API_KEY you provide.
Does it support remote/OAuth?
It runs locally over stdio with npx and a TURBOPUFFER_API_KEY. turbopuffer also hosts a remote Streamable HTTP endpoint at https://turbopuffer.stlmcp.com, where the key is passed via the x-turbopuffer-api-key header or an Authorization: Bearer token. There is no OAuth flow.
How does it expose write and query operations?
Rather than one tool per endpoint, it ships an execute tool that runs TypeScript against the official turbopuffer SDK in a sandbox, so the agent can write, query, delete, and inspect documents directly, plus a search_docs tool to ground itself in the API.
← Browse all vector-search servers