Openzim MCP

What cameronrye/openzim-mcp MCP server does

The cameronrye/openzim-mcp MCP server gives MCP-compatible AI clients access to local ZIM files, a compressed archive format used for offline knowledge collections. It is suited to archives from Wikipedia, Wiktionary, Stack Exchange, and other Kiwix Library sources. The server supports structured retrieval rather than exposing an archive as an undifferentiated text dump.

In Simple mode, the server presents zim_query, which accepts a natural-language request and selects the underlying operation. Advanced mode exposes eight tools: zim_query, zim_search, zim_get, zim_get_section, zim_browse, zim_metadata, zim_links, and zim_health.

Retrieval includes full-text search, suggestions, pagination, article and section access, namespace-aware browsing, archive metadata, and link-graph queries. The server also supports inbound-link discovery, allowing an agent to find content that links to a given entry. Archive-type presets adjust retrieval behavior for sources such as Wikipedia and Stack Exchange.

How it works

The cameronrye/openzim-mcp MCP server runs locally against one or more ZIM archives supplied through a filesystem directory. Native libzim support is used for archive inspection and validation. Caching and paginated operations are included for work with large archives.

A local stdio transport is available for MCP clients. The project also supports Streamable HTTP for a long-running service, with bearer-token authentication, CORS, and health endpoints. Each archive entry can be exposed as an MCP resource using a zim:// URI. Resource subscriptions can notify clients when an archive appears, disappears, or is replaced.

Setup and configuration

Install the Python package with either uv tool install openzim-mcp or pip install openzim-mcp. A multi-architecture Docker image is also published at ghcr.io/cameronrye/openzim-mcp. The server needs a directory containing .zim files; it does not provide an archive itself.

A basic local launch uses openzim-mcp ~/zim-files. For a client configuration, the README shows uvx with openzim-mcp and the archive directory as arguments. Advanced mode adds --mode advanced before the directory. The documented Claude Desktop configuration uses stdio and works with clients that support that transport.

HTTP deployment is enabled with OPENZIM_MCP_TRANSPORT=http, along with host and authentication-token settings shown in the project documentation. Preset defaults can be overridden through OPENZIM_MCP_PRESETS_OVERRIDE_PATH, which points to a TOML configuration file.

Limitations and notes

The cameronrye/openzim-mcp MCP server cannot answer from an archive that has not been downloaded and placed in its configured directory. Full ZIM collections may range from hundreds of megabytes to tens of gigabytes, so local storage and archive access are part of deployment planning. The README documents local stdio and optional HTTP operation; it does not describe a hosted archive service or built-in live web retrieval.