XWiki MCP Server: Connect AI agents to your wiki while keeping control over your data

I'm part of the XWiki team, and we've recently released an open-source MCP server that lets AI agents interact directly with content stored in XWiki, our knowledge management and collaboration platform.

The idea came from a challenge we're seeing more often: organizations have years of documentation, procedures, and structured knowledge in their wikis, and they want to make that information accessible to AI agents without moving it to another platform.

With the MCP server, agents like Claude Code, Codex, and OpenCode can search and read wiki pages, access attachments and page history, query structured data, and create or edit content.

One design decision we'd particularly like feedback on is how we handle permissions and write access. The MCP server uses XWiki's existing permission system, so authenticated agents act with the rights of the user they're connected as. Write operations are disabled by default and need to be explicitly enabled by an administrator.

The extension is optional and self-hostable. It doesn't require a particular AI provider, so organizations can decide which agents and models they use. Where data is processed still depends on that choice.

There's some setup involved, including configuring authentication and connecting your MCP client to your XWiki instance.

Source code: https://github.com/xwiki-contrib/ai-llm

Documentation: https://www.xwiki.org/xwiki/bin/view/documentation/extensions/admin/llm/mcp-server/

We also launched the project on Product Hunt today:
https://www.producthunt.com/products/xwiki/launches/mcp-server-by-xwiki

For those already experimenting with MCP and internal knowledge bases, how are you handling permissions and write operations? Do you prefer keeping agents read-only by default, or do you find that too restrictive for practical workflows?

We'd appreciate feedback, especially from people who have tried connecting AI agents to existing documentation systems.