Every MCP operation runs as the authenticated Confluence caller. Native permissions and MCP policies determine what that caller can do; tools cannot override identity, permissions or policy.
Authentication and tokens
Use HTTPS. Both OAuth and personal access tokens (PATs) use your existing Confluence account.
|
Method |
Access |
|---|---|
|
OAuth |
Requires administrator-enabled MCP OAuth and explicit user consent. Granted |
|
PAT |
Uses the token owner’s native Confluence permissions, including write access where permitted, subject to MCP policies. No MCP OAuth registration or consent is required. Disabling MCP OAuth does not disable PAT access. |
Basic or session authentication depends on the client and site configuration. All accepted MCP endpoint URLs enforce the same authentication and scope checks. Use confluence_get_current_user to verify the account; see Identity tool and result fields.
Access enforcement
The caller must pass the MCP audience, mode, space policy and any OAuth scope checks. These controls apply to administrators too and never grant native permissions.
-
Audience: determines who may use MCP, including discovery, diagnostics and existing transfers or cursors.
-
Writes: Read-only mode and
mcp.readtokens hide and refuse content-changing tools. MCP mode does not change native permissions outside MCP. -
Space policy: Native permissions adds no filter; Allowlist narrows eligible spaces and Blocklist excludes spaces.
Administrator setup: Configure MCP policies.
Permissions by operation
Current page and space access is required, including for related content and history. Missing, restricted or policy-excluded content is not disclosed; a denial does not establish that hidden content exists.
|
Operation |
Native requirement and effect |
|---|---|
|
Read, search and navigate |
Space/page view access; returned pages and every navigation path edge are checked. Hidden totals and paths are not disclosed; personal spaces follow the same rules. Search follows Confluence’s indexing schedule, so recent changes may be readable by ID before appearing in search. |
|
Create pages |
Space view and page-create permission. An optional parent must be visible in the target space. |
|
Edit pages or sections |
Page view/edit permission and current version. Retains space, ancestry and restrictions; Confluence records the caller’s authorship. Section selection and storage rules are in Editing contracts. |
|
Comments |
Current page access plus native comment-create, edit or remove permission, including author-specific rules. Page edit permission is not a substitute. Mutations support footer comments. |
|
Labels |
Page edit and native label-action permission. Only the named global label changes; unrelated and personal/system labels remain. |
|
History |
Current page/space access. Historical content grants no broader access; restoring a historical revision is unavailable. |
|
Attachments |
Current owning-page and attachment visibility. Upload/version requires native attachment-create/update permission and site restrictions; copies check source and destination access and both space policies. |
|
Page/tree copy |
Every source and ancestor must be visible and unrestricted; destination view/create permission. Inaccessible descendants are refused. Copies inherit destination access. |
|
Move/reorder |
Native move permission, source edit/destination create rules and bounded-subtree access. Same-space only. Identity and direct restrictions remain; inherited access follows destination ancestry. |
|
Trash |
Native page-remove permission and trash validation. Current non-homepages without children only; reversible, with no recursive trash or permanent purge. |
|
Trash discovery/restore |
Native space administration, visibility and restoration permission. Restore needs a current parent in the original space; title, child, state and destination-access conflicts are refused. |
Client and data handling
Retrieved content reaches the MCP client and any AI service it uses. Choose a client appropriate for your organization’s content and treat retrieved pages as source material, not instructions. Client tool toggles and approval prompts do not change server permissions.
Store tokens in client secret storage, set an appropriate expiry and revoke them when no longer needed. Keep credentials out of prompts, shared files and source control.
App diagnostic events omit usernames, credentials, query text and page content. This does not guarantee redaction in other Confluence, proxy or client logs. See Diagnostics for request references and logging controls, and Support and security reports for reporting. Share only sanitized information.