MCP Server by XWiki
The interesting part is the control angle. XWiki gives you a bridge where the AI can query your knowledge but you decide what it sees and what it can touch. That matters because the biggest fear I hear from teams about AI is losing grip on their own data, especially in regulated or internal-knowledge-heavy environments.
I think this is the right instinct. Wikis are still the messy, organic home of a lot of company knowledge, and they've been under-appreciated in the AI gold rush. Everyone's racing to build RAG pipelines on top of clean data, but most real knowledge lives in pages written by ten different people over five years. MCP is becoming the connective tissue for this stuff, and XWiki showing up with a proper server means the open-source wiki crowd doesn't get left behind.
What I'd want to know next is how granular the permissions are and whether it plays nicely with existing access controls. If the AI only surfaces what a user is already allowed to see, that's a real differentiator. If it's a blunt instrument, it's just another integration. The control promise is the whole pitch, so the execution has to match.