Search and inspect
Search documents, retrieve records, list collections, and inspect schemas through validated tool inputs.
Model Context Protocol
Typesense MCP server for AI agents
Connect Claude, Codex, Cursor, and other MCP clients to Typesense through structured, typed tools—read-only by default.
TypesenseKit MCP is an independent, community-built open-source server. It shares the same operation registry as the TypesenseKit CLI, so agents and operators use consistent inputs and outputs.
MIT licensedRead-only defaultStdio + HTTP
Run with pnpm
TYPESENSE_URL=https://your-cluster.typesense.net \
TYPESENSE_API_KEY=your-scoped-api-key \
pnpm dlx @typesensekit/mcpNo global install or server runtime is required. Use a narrowly scoped Typesense API key.
One MCP surface
Give agents useful Typesense context
Let an assistant inspect a schema, search documents, check cluster health, and explain structured results without handing it an unbounded shell script.
Safe agent access
Write, delete, key-management, and raw API tools stay hidden until full access is deliberately enabled.
Structured results
Agents receive predictable JSON-shaped results from the same typed operations available in the CLI.
Supported clients
Use Typesense with your AI coding tools
Copy supported configurations for Claude Desktop, Claude Code, Codex, Cursor, or any client that can launch a stdio MCP server.
View client configurations →Read-only by default
Start with search and inspection tools. Enable full access only for a controlled workflow.
Scoped Typesense keys
Combine the server default with a narrowly scoped API key for defense in depth.
Stdio or Streamable HTTP
Run locally through stdio or follow the production guidance for authenticated remote access.
Common questions
Typesense MCP FAQ
Installation, compatibility, ownership, and access defaults at a glance.
What is a Typesense MCP server?
A Typesense MCP server connects Model Context Protocol clients to a Typesense cluster. TypesenseKit gives AI agents structured tools for search, schema inspection, collection metadata, administration, and other supported Typesense operations.
Is TypesenseKit an official Typesense project?
No. TypesenseKit is an independent, community-built MIT-licensed open-source project for working with Typesense.
Is the Typesense MCP server read-only?
Yes by default. Search, reads, collection metadata, configuration reads, and system status tools are available, while write, delete, key-management, and raw API tools require an explicit full-access opt-in.
Which AI clients work with TypesenseKit MCP?
TypesenseKit supports stdio configurations for Claude Desktop, Claude Code, Codex, Cursor, and generic MCP-compatible clients. It also supports Streamable HTTP deployments.
How do I install the Typesense MCP server?
Run it without a global install using npx -y @typesensekit/mcp or pnpm dlx @typesensekit/mcp, with TYPESENSE_URL and TYPESENSE_API_KEY in the environment.
Open source
Bring safe Typesense tools to your agent.
Try the MCP server, inspect the code, and star TypesenseKit if it improves your search workflow.