OrgMCP/ docs
Sign up
Documentation
Get started
Connectors
MCP integration
Reference
Enterprise
Misc

Confluence connector

Two flavours: Atlassian OAuth (recommended — one consent also installs Jira) and email + API token (manual). Both power the live MCP tools (confluence_list_spaces, confluence_get_page, confluence_search, confluence_diagnose) and can index a space into the knowledge corpus as confluence_space.

Option A — Atlassian OAuth (recommended)

  1. In the OrgMCP admin console, open Connectors → Add → Confluence (or Jira — the same start URL fans out into both connector rows).
  2. Click Connect to Atlassian. Approve the read scopes (Confluence content/space/user + Jira work/user + offline_access).
  3. Atlassian redirects to /api/oauth-tenant/atlassian/callback. We persist the token blob (including refresh token + client id/secret) and bounce you back with both Confluence and Jira rows.
  4. On the connector's Sources tab, pick the spaces to index as knowledge sources.

Ops note: register https://staging.app.orgmcp.io/api/oauth-tenant/atlassian/callback and https://app.orgmcp.io/api/oauth-tenant/atlassian/callback on the Atlassian 3LO app, and populate ATLASSIAN_OAUTH_* Terraform vars so the Connect CTA enables.

Rotation: access tokens expire in ~60 minutes and refresh automatically when a refresh_token is present. Reconnect if refresh fails.

Option B — Email + API token

  1. Create an Atlassian API token, then paste Host, email, and token in Connectors → Add → Confluence.
  2. Click Test connection. A green result requires at least one visible space; zero spaces fails (use confluence_diagnose / the diagnostics panel).

What gets indexed

  • Published pages and blog posts in each selected space (storage HTML stripped to plain text). Footer + inline root comments (up to 40 per page/blog) are appended into the parent document. Full re-crawl on the Confluence cadence.
  • Shared document budget: pages are crawled first; blog posts use whatever remains of the 1000-document space cap.

What does not get indexed

  • Attachments (binary / MIME / OCR — not in RAG yet).
  • Nested comment replies beyond the root footer/inline comments returned in the first page.
  • Spaces the credential cannot see. Diagnose surfaces the visible count; Test fails at zero.

Honesty surface

  • confluence_diagnose / admin Test — mode, host, visible space count, sample keys. Zero spaces ⇒ test fails.
  • List / search tools paginate with pageToken / nextPageToken.
  • Partial syncs at the page cap write a warning; empty crawls write a precise emptyReason.

See also Jira connector (same Atlassian grant) and the canonical matrix in the operator docs (docs/ARCH-connector-capabilities.md).