Set up the Searchcraft MCP Server

The Searchcraft MCP Server connects AI clients like Claude Code or Claude Desktop to your own Searchcraft search instance, making index management, document import, and search queries controllable through natural language. This guide covers setup, requirements, and a safe first test run.

Set up the prerequisites

Before the MCP server can be used, a running Searchcraft instance is required — either a cluster in the managed Searchcraft Cloud (Vektron Dashboard) or a self-hosted Searchcraft Core installation. From there you need the endpoint address and an API key. Node.js 22 or newer is also required.

Build and connect the server

After cloning the searchcraft-inc/searchcraft-mcp-server repository, yarn installs the dependencies and yarn build produces the stdio and HTTP variants of the server. Store the credentials (ENDPOINT_URL, CORE_API_KEY) in a .env file. For Claude Code, this is then enough: claude mcp add searchcraft -- node /path/to/dist/stdio-server.js. Claude Desktop instead wires up the server through an entry in claude_desktop_config.json.

Keep the API key narrowly scoped

Because an agent operates with the permissions of the configured key, it is worth using the create_key tool to generate a dedicated key with permissions scoped as narrowly as possible to the task at hand, rather than reusing an existing full-access key.

Create and test a first index

A safe first test is creating an index from a public sample JSON file with create_index_from_json, followed by a simple query via get_search_results. Anyone who also wants to see an interface can generate a standalone React search app with create_vite_app.