The two-step search-then-fetch shape is deliberate: search returns 200-character previews so a broad query stays cheap, and you only pull full bodies for the hits that matter. The workflow tools are the less obvious half — an assistant can pick up a process assigned to you and carry it forward rather than just describing it.
An MCP server for Runbook, a documentation and workflow tool. It lets an assistant search the knowledge base, read and write articles in Markdown, and run the workflow processes assigned to you — the same books and processes your team already uses.
- runbook-search-articles — keyword search across books and workspaces; bodies come back truncated to 200 characters
- runbook-get-article — the full article body by ID or URL, which is how you follow up on a search hit
- runbook-list-articles and runbook-list-books — the top 100 articles in a book, and the books in the organization
- runbook-create-article and runbook-update-article — write Markdown into a book, with category IDs attached
- runbook-list-categories and runbook-get-folder — categories in a book, and a folder with its ancestors, subfolders and articles
- runbook-list-assigned-processes — the in-progress workflow runs waiting on you
- runbook-get-process and runbook-run-process — read the current state of a workflow, and start a new run or continue an existing one
- runbook-upload-run-state-file — a presigned URL for attaching a file to a running process
A Runbook API token and your instance URL, passed as RUNBOOK_API_TOKEN and RUNBOOK_BASE_URL (https://<your-subdomain>.runbook.jp). Clients launch it with npx -y @runbook-docs/mcp-server over stdio and need Node.js 20 or newer; Claude Desktop can instead install the .mcpb bundle from the releases page. Cursor takes the same values as --api-token and --base-url flags.
One command plus a key — npx -y @runbook-docs/mcp-server, then supply credentials