confluence-dc-advops-mcp

Local MCP helpers for Confluence Data Center used from Cursor.

Русская документация: README.ru.md.

Talks to the same proxy/auth as @atlassian-dc-mcp/confluence (local TLS proxy + token from keychain / env). Prefer these tools over ad-hoc curl to the public Confluence hostname.

Auth

Same sources as @atlassian-dc-mcp/confluence:

• CONFLUENCE_HOST (env or ~/.atlassian-dc-mcp/confluence.env) — typically your local proxy, e.g. https://localhost:8443

• CONFLUENCE_API_TOKEN (env), or macOS Keychain service atlassian-dc-mcp / account confluence-token

Do not commit tokens or *.env files. See SECURITY.md.

Related MCP server: confluence-mcp-server

Tools

Fast path for large pages (BRD/SRS/…)

1. confluence_getStorageToFile → /path/to/page.xml

2. confluence_storage_listHeadings / getSection (format: text) — do not @ or Read the dump file into chat

3. confluence_storage_replaceSection or confluence_storage_replaceMacroBody (dryRun first if unsure)

4. confluence_updateStorageFromFile (omit version to auto-bump, or pass current+1)

5. Verify with user-confluence-dc / confluence_getContent bodyMode: text (limit chars)

6. If the page is also a Create-from-template snapshot → sync space template (below)

7. Delete the temp file

These storage tools do not publish and do not return the full XML. Python/StrReplace on the dump is a fallback when they do not cover the case.

Page versions (who / when — not full blame)

1. confluence_listVersions — metadata only (number, when, message, by). Paginate with start / limit; cap with maxResults.

2. To inspect an old body: confluence_getStorageToFile with version: N → local XML, then storage_getSection / search on disk. Do not pull historical XML into chat.

3. Not git-blame: a version author is who saved that snapshot; use snippet search on the dump for “who wrote this line”.

Do not round-trip the whole page through Markdown (macros/layout will not survive).

For tiny pages that are not templates, user-confluence-dc confluence_updateContent is still ok.

Links in text/markdown

Confluence often stores page links as empty ac:link with the target only in ri:* attributes (UI fills the title). format: storage is unchanged. For format: text / markdown (and headings that are only a link), the converter fills a label so agents do not treat the field as empty:

1. Prefer visible anchor text (ac:plain-text-link-body / ac:link-body); do not also append ri:content-title.

2. Otherwise first target: ri:page / ri:blog-post → ri:content-title plus [spaceKey] when ri:space-key is set; ri:url → ri:value; ri:attachment → ri:filename; ri:space → key/name; ri:user → [user] (no invented display name).

3. Markdown: same labels; ri:url may become [label](url). Page links stay bare titles (no invented /wiki/… URLs).

4. No REST lookup of titles/ids; no default space when ri:space-key is missing; tiny/pageId not resolved here.

Same contract as user-confluence-dc confluence_getContent with bodyMode: text (upstream mapper). Non-expand macros still appear as [macro: …] stubs in getSection text — bodies inside those macros are not walked. ac:inline-comment-marker (yellow highlight) is unwrapped: inner text is kept, ac:ref is not.

Tiny URLs (/x/{code})

DC encodes pageId as little-endian uint32 → standard base64, then /→- and +→_ (Atlassian KB). This is not RFC4648 base64url. confluence_resolveTinyUrl:

1. Decode with the DC alphabet → GET /content/{id}.

2. On 404, retry decode with -↔_ swapped (codes produced by a base64url encoder).

3. If still 404, follow pages/tinyurl.action?urlIdentifier={code} (undocumented; CONFSERVER-102023).

4. Response includes resolvedVia: decode | decode-swapped | tinyurl-action.

Use before getContent / getStorageToFile when the user pasted a short link — not CQL by title. Anchors after the code (/x/ZIAB#fragment) are ignored.

Fast path: page → space template (Create from template)

Catalog / BSA page = source of truth. Space template = snapshot for «Create from template».

confluence_syncPageToSpaceTemplate
  contentId: "<pageId>"
  spaceKey: "<spaceKey>"
  templateId: "<templateId>"
  descriptionSuffix: "(sync note)"   # optional
  # copyPageLabels: true by default — template labels come from the page

DC notes:

• List: /rest/experimental/template/page?spaceKey=…

• GET by template id alone often 404 — always pass spaceKey

• Update: PUT /rest/experimental/template with templateType: "page" and body.storage

• Delete: DELETE /rest/experimental/template/{id} — only via MCP delete tools after human confirmation (confirm: "DELETE" + exact confirmName / confirmNames). No trash restore.

Labels (pages + Create from template)

Confluence copies space-template labels onto pages created from that template. Put labels on the catalog page, then sync (or set them on the template). New pages from «Create from template» inherit them.

confluence_syncPageToSpaceTemplate copies the page’s global labels onto the template (copyPageLabels default true). If the page has none, current template labels are kept. Pass labels: [...] to set them explicitly. Create/update-from-file accept labels and copyLabelsFromContentId.

Page API: GET/POST /rest/api/content/{id}/label, DELETE …/label?name=. Template labels go in the experimental template PUT payload (omitting them can wipe labels on DC — tools always send a list unless you set keepLabels: false with no replacement).

Footer comments (page comments)

No inline comments. Put a quote from the page in the comment body so the subject is clear.

bodyFormat: plain (default — wrap in <p>, blank lines = paragraphs) or storage (raw Confluence storage XML).

Inline comments (read + reply only)

Does not create new text anchors. Use footer comments with a quote if you need a new note.

Each item includes status (open|resolved), visibleOnPage (true when open on the page), and originalSelection (anchored text when present).

Cursor config

"confluence-dc-advops": {
  "command": "node",
  "args": ["/path/to/confluence-dc-advops-mcp/index.js"],
  "env": {
    "CONFLUENCE_HOST": "https://localhost:8443",
    "NODE_TLS_REJECT_UNAUTHORIZED": "0"
  }
}

After changing index.js, reload MCP servers in Cursor so new tools appear.