MCP SERVERSearch a corporate XWiki, read a page with its attachments and child pages, and write back to it — create_page upserts, add_comment records a note beside the page instead of inside it.
LABSCO SUMMARY
create_page is an upsert: writing to a space and page address that already holds a page replaces it, so updating a document and replacing one by accident are the same gesture at the call site. delete_page's own description says an administrator has to restore a page afterwards, and reserves the call for an explicit deletion request rather than tidying up. On the read side everything is addressed the same way — get_page, get_attachments and get_page_children all take space plus page, with dot notation for nested sections — which makes search the practical starting point, since it hands back the addresses those calls need.
A server over a corporate XWiki instance, exposing search, space and page browsing, attachment listing, child-page traversal, page creation and deletion, and commenting.
A reachable XWiki instance, whose base URL you supply as XWIKI_BASE_URL.
One command plus credentials — npx -y xwiki-mcp, then supply credentials
USE CASES
Search the wiki for the term, then open the page the search points at and read it as written.
List the top-level sections, look at what is inside the one that matters, and follow an index page down to its sub-articles.
Read the page, then list what is attached to it and take the download URL for the file that holds the actual data.
Pick the space it belongs in, then write the page — the same call creates a new one or replaces what is already at that address.
Read the page, then leave the correction as a comment underneath it so the body stays as its author left it.
Counted once with only startup setup. Other versions, modes or optional credentials may expose a different list.
add_commentPost a comment underneath a wiki page (appears in the page Comments section). Use this when:
Do NOT use this for substantive content — that belongs in the page body via create_page. Returns: {ok: true}.
create_pageWrite a page to the wiki. Creates a new page OR updates an existing one (same call — XWiki treats it as upsert). Use this when the user says: "save to wiki", "publish", "write up", "document this", "create a page", "запиши в вики".
WHERE TO PUT IT — pick a sensible location:
CONTENT FORMAT — XWiki markup (NOT Markdown by default):
Returns: {ok, title, version, url}. The url is the human-facing page link.
delete_pageDESTRUCTIVE: delete a page from the wiki. Only an admin can restore it afterwards.
Use ONLY when the user explicitly says: "delete the page", "remove X from the wiki", "удали страницу".
Do NOT call this tool to "clean up", "refactor", or as a side effect of any other request.
If the user asks to "update" or "rewrite" a page, use create_page instead (it overwrites).
Returns: {ok, deleted: "Space.Page"}.
get_attachmentsList files attached to a wiki page (screenshots, PDFs, CSVs, etc.). Use this when:
Returns: array of {name, size_bytes, mime_type, author, date, download_url}. NOTE: this tool does NOT download file contents — use the download_url separately if needed.
get_pageRead the full content of one wiki page (the actual text, formulas, links, images). Use this when:
search returned a relevant result and you want to read it — pass space and page straight from the search resultIMPORTANT — how to address a page:
space is the FULL path of containing folders joined with dots. Example: "Documentation.API.v2"page is just the leaf page name. For section home pages this is usually "WebHome".Returns: {title, content, syntax, author, modified_date, version, url}. content is XWiki markup (typically xwiki/2.1) — not Markdown; "= H1 =" is a heading, "[[label>>target]]" is a link.
get_page_childrenList direct child pages of a given parent page (one level deep — does not recurse). Use this when:
Returns: array of {id, title, parent, url}. NEXT STEP: call get_page on a specific child, or recurse with get_page_children to go deeper.
list_pagesList pages inside a wiki space. Returns titles + URLs for browsing. Use this when:
search returned a space-level hit and you want to dig deeperFor nested sections: use DOT notation. Example: space="Documentation.API" reads /Documentation/API/. Returns: {pages: [{id, title, parent, url}], _pagination: {...}}. NEXT STEP: pick a page and call get_page with the same space + the page name.
list_spacesList all top-level sections ("spaces") of the corporate XWiki. A "space" is like a folder — e.g. "Documentation", "Sandbox", "AI". Use this when:
search returned nothing and you need a fallback way to browselist_pages or get_page.Returns: array of {id, name, home_url}. NEXT STEP: pick a relevant name and pass it to list_pages to see what is inside.
searchPRIMARY ENTRY POINT for the corporate XWiki. Use this FIRST whenever the user mentions:
DO NOT guess page paths. DO NOT call get_page blindly. Search first, then read.
This is a real full-text search over page CONTENT (Solr-indexed), not just titles. Pages are often in Russian — just pass the user query as-is, do not translate.
Returns ranked results: each item has {id, title, space, url, score, modified_date}. A response also includes the actual engine used and _pagination.
WORKFLOW: search → read top results with get_page (use the space and the page name from the result id) → if the right page is missing, fall back to list_spaces / list_pages for manual browsing.
BEFORE YOU INSTALL
9 tools when we connected.
Both documented clients run the server on your machine.
CONTEXT COST
Rated by the size of the 9 tool definitions that ride along on every turn, not by how many there are
INSTALLATION
npx -y xwiki-mcp{
"mcpServers": {
"xwiki": {
"command": "npx",
"args": [
"xwiki-mcp"
],
"env": {
"XWIKI_BASE_URL": "<YOUR_XWIKI_BASE_URL>",
"XWIKI_AUTH_TYPE": "basic",
"XWIKI_USERNAME": "<YOUR_XWIKI_USERNAME>",
"XWIKI_PASSWORD": "<YOUR_XWIKI_PASSWORD>"
}
}
}
}MCP servers
anuragagrawal0430
makenotion
mstang
Vortiago