Searchcraft MCP Server

The Searchcraft MCP Server is the official Model Context Protocol connector from Searchcraft Inc., a US-based provider of a Rust-built, developer-first search engine. According to the official repository at https://github.com/searchcraft-inc/searchcraft-mcp-server (Apache-2.0 license, 9 stars recorded via the GitHub API on 2026-09-19, actively maintained, package.json version 0.2.0), the server connects MCP-capable clients such as Claude Desktop or Claude Code directly to a running Searchcraft instance and lets you manage an entire search engine purely through natural-language instructions instead of writing code or operating a dashboard.

What is Searchcraft, and what is the MCP server for?

Searchcraft positions itself, according to the provider, as an alternative to classic search solutions like Elasticsearch or costly vector-database setups: rather than relying solely on semantic vector search, the engine combines high-performance keyword search with an additional "semantic control layer." According to the provider, the target audience is primarily developer teams building product search for B2B SaaS applications, searchable documentation, publishing platforms, or retrieval for generative AI and agentic workflows. The MCP server makes this functionality usable for AI coding assistants: an agent can create indexes on the developer's behalf, import sample data, run search queries, and even generate a ready-made search interface, without anyone having to call the Searchcraft API by hand.

Features and tools

According to the README, the server exposes over 50 tools across three categories. The "Engine API tools" cover complete management of a Searchcraft instance: creating, configuring, and deleting indexes; adding, updating, or removing documents; generating access keys with granular permissions (the create_key tool); maintaining synonyms and stopwords; and running search queries plus retrieving analytics (get_search_results). The "import tools" lower the barrier to entry: create_index_from_json automatically derives a matching index schema from a JSON file or URL and populates it, without anyone having to model the data structure by hand first. The third category, app generation, stands out in particular: the create_vite_app tool turns existing index data into a complete, standalone search application built on Vite and React, including a read-access key — letting an agent go from "raw data as JSON" to a working, searchable web interface in a single pass.

Deployment modes and requirements

The server itself can be run in several ways: as a stdio process for Claude Desktop and Claude Code (the recommended path per the README), as a local HTTP server for testing, or as a Docker container for remote or server deployment. In every case, though, the server requires an already-existing Searchcraft instance to talk to: the environment variables ENDPOINT_URL (the Searchcraft cluster address) and CORE_API_KEY (the API key) must be set. That instance can be either a managed Searchcraft Cloud environment (Vektron Dashboard) or a self-hosted "Searchcraft Core" server. Node.js 22 or newer is required to build and run it.

Pricing model: free connector, paid engine

The MCP server's source code itself is Apache-2.0 licensed and therefore free to use. The underlying Searchcraft search engine, however, is not a free open-source engine according to the provider: the official FAQ states plainly, "open-source software is free, running search itself is not." According to the provider, production use of self-hosted Searchcraft Core requires a license and support agreement with Searchcraft; alternatively, the managed Searchcraft Cloud offering is available with usage-based billing. The public website does not disclose specific amounts, directing interested parties to a contact form for custom quotes. Anyone wanting to try the MCP server therefore needs credentials to a running Searchcraft instance from the outset.

Who is the Searchcraft MCP Server a good fit for?

The server is best suited to teams already working with Searchcraft, or evaluating it as a search solution, who want to bring AI coding assistants into the setup and maintenance process — for example, to quickly spin up prototype indexes from existing JSON exports or demonstrate an initial searchable interface. Anyone looking for a purely free, open-source search engine without vendor lock-in will find other permanently free alternatives for related data-access and research tasks in the Skill Road catalog, such as the Redis MCP Server or Fetch MCP.

FAQ

Is the Searchcraft MCP Server free? The connector's source code is Apache-2.0 licensed and free. Operating it, however, requires a Searchcraft instance, and according to the provider, production use of that instance is paid (cloud billing or a self-host license agreement).

Do I need a Searchcraft account? Yes. Without an ENDPOINT_URL and CORE_API_KEY for a running Searchcraft Cloud or self-hosted Core instance, the server does not function.

What does create_vite_app do? The tool automatically generates a complete React application built with Vite from existing index data, including a read-only access key for the new search app.

How current are the GitHub stars? The GitHub API reported exactly 9 stars for the repository https://github.com/searchcraft-inc/searchcraft-mcp-server on 2026-09-19. That's a point-in-time snapshot, not a quality judgement.

Node.js 22 or newer; an already-running Searchcraft instance (cloud or self-hosted Core) with an endpoint URL and API key; an MCP-capable client with support for local stdio servers.

Clone the repository, install with yarn, and run yarn build. Create a .env file with ENDPOINT_URL and CORE_API_KEY for your own Searchcraft instance. For Claude Code: claude mcp add searchcraft -- node /path/to/dist/stdio-server.js. Alternatively, build the Docker image and start it with the same environment variables.

claude mcp add searchcraft -- node /path/to/dist/stdio-server.js

API key authentication against your own Searchcraft instance via the CORE_API_KEY environment variable; keys can be generated with granular permissions using the create_key tool.

Full read and write access to the configured Searchcraft instance according to the permissions of the API key in use: creating, modifying, and deleting indexes and documents, plus running search queries and analytics.

Requests go from the locally run MCP server directly to the configured ENDPOINT_URL of your own Searchcraft Cloud or self-hosted Core instance. No data passes through any additional third-party servers operated by Searchcraft beyond the chosen cloud cluster itself.

Depending on configuration, the API key grants broad write permissions on the Searchcraft instance (creating/deleting indexes, modifying documents); keys should therefore be generated with the narrowest possible granular permissions and never shared in plain text. Since the server runs locally, there are no additional risks from a third-party-hosted intermediary service.