Worth knowing before you plan around it: the Wiki.JS GraphQL API has no native pagination, so paging is simulated on the client side, and neither search nor list results carry page bodies. Budget two calls — find the page, then fetch it by ID or path for the content.
A TypeScript MCP server that connects to your own Wiki.JS instance and lets an assistant find and read pages from it. It talks to the wiki's GraphQL endpoint and handles the query construction for you.
- `search_wiki_pages` — keyword search with page and limit arguments for paging through results
- `list_wiki_pages` — a paginated listing of everything available
- `get_wiki_page` — fetch one page by numeric ID or by path, and this is the call that returns full content
- A connection test you can run before wiring it into a client
- Two run modes — stdio, or HTTP with a /health check and a /mcp endpoint
- A published Docker image with a healthcheck, and a compose example
A running Wiki.JS instance and an API key whose permissions include page:read, page:list and search:read. Set WIKI_JS_BASE_URL and WIKI_JS_API_KEY, plus optional LOG_LEVEL, MCP_MODE and PORT — either in a .env file or as system environment variables. Install with npm, build, and start; or run the Docker image. MIT licensed.
One command plus a key — docker run -d \ --name avr-docs-mcp \ -e WIKI_JS_BASE_URL="https://your-wiki-instance.com" \ -e WIKI_JS_API_KEY="your-api-key-here" \ agentvoiceresponse/avr-docs-mcp:latest, then supply credentials