Confluence MCP Server

The Confluence MCP server provides 6 tools to search, read, create and update Confluence pages, list child pages and add comments. It works with Confluence Cloud and with Confluence Data Center / Server.

Setup

Confluence Cloud and Data Center use different credentials.

1. Get the credentials:
   • Cloud (*.atlassian.net): an Atlassian API token and the email of its Atlassian account.
   • Data Center / Server: a personal access token, created in your Confluence profile.
2. In FlowTrux, go to Resources > MCP Servers > Add Server (OWNER or ADMIN) and choose Confluence.
3. Give it a Name and fill in:

4. Under Available in workspaces, pick the workspaces, then click Add Server.

A 401 means the credentials do not match the deployment (for example an API token used against Data Center). A 404 on every call on Cloud usually means /wiki is missing from Base URL.

Workflows act as the authenticated user: a page that user cannot view answers 404, one they cannot edit answers 403.

Tools

Finding pages

search_pages

Full-text search for pages. query is words to find; it may also be a CQL expression (title ~ "onboarding" AND label = "hr") when you need more than text matching. Returns id, title, space, url and an excerpt per hit.

list_child_pages

Direct child pages of a page (one level), in Confluence's tree order.

search_pages takes plain words, or a CQL expression when text matching is not enough: title ~ "runbook" AND label = "ops", ancestor = 12345 for a whole subtree, lastmodified > now("-7d"). It returns page ids; the other tools take a pageId.

Reading and writing

get_page

Read one page: title, space, version, URL, parent chain, and the body converted to readable text.

create_page

Create a page in a space. body is plain text or light Markdown (converted for you); storage format is accepted as is. The title must be unique within the space.

update_page

Replace a page's body (and optionally its title). The current version is fetched and incremented for you. This replaces the whole body: read it with get_page first if you mean to append.

add_comment

Add a footer comment to a page. body is plain text or light Markdown.

Page bodies are converted both ways. get_page returns readable text: # headings, - bullets, fenced code, pipe tables and [text](url) links; macros keep their text and images appear as [image: file]. create_page, update_page and add_comment take plain text or light Markdown (blank line = new paragraph, #, -, 1., code fences, **bold**, links). A body that already is Confluence storage format is sent unchanged.

update_page replaces the whole body. To append, read the page with get_page and send the old text plus the addition. Pages heavy with macros can lose macro settings on such a round trip - edit those in the browser.

Titles are unique per space. create_page fails with A page with this title already exists when the title is taken - find the page and use update_page. space is the space key (ENG), not its name. A page without parentId lands at the top of the space. Comments created here are page (footer) comments, not inline comments.

Examples

Publish a weekly status page:

Schedule → Action:MCP(jira, search_issues, {jql: "project = OPS AND updated >= -7d"}) → Agent (write the title and summary) → Action:MCP(confluence, create_page, {space: "OPS", title: "{{steps.agent-1.output.response.title}}", body: "{{steps.agent-1.output.response.summary}}", parentId: "12345"})

Answer questions from the wiki:

Webhook (question) → Agent (tools: confluence:search_pages, confluence:get_page) → reply with the answer and the page links

Append to a runbook:

Action:MCP(confluence, get_page, {pageId: "67890"}) → Transform (old body + new entry, as `text`) → Action:MCP(confluence, update_page, {pageId: "67890", body: "{{steps.merge.output.text}}", versionMessage: "Added incident 42"})