The win is the round trip: find the page with a CQL search, read it as markdown, and write the edit back with the original markup intact — without opening a browser tab or hand-converting storage format.
An MCP server for Confluence. It authenticates with a personal API token, fetches pages and spaces through the REST API, and cleans the content on the way out so a model reads prose rather than XHTML.
- A page fetched by ID, returned as text or as markdown
- The original Confluence Storage Format markup alongside it, so an update can preserve the existing formatting
- Search across pages with CQL, the query language the Confluence UI itself uses
- The spaces your account can reach, listed
- New pages created, and existing pages updated
- Comments read and added
- Attachments listed and uploaded
Bun v1.0.0 or higher — the server is built and run with Bun, not Node. Three environment variables: a personal API token, your instance URL ending in /wiki, and the email address that token belongs to. There is no published package to npx: clone the repository and `bun run build` before pointing a client at `dist/index.js`.
One command plus a key — npx -y confluence-mcp, then supply credentials