What happens to your indexed data when Mistral flips the switch?

Mistral has set a hard deadline. By August 31, the Google Drive and SharePoint Knowledge Connectors that Vibe Work customers currently use will be shut down and their indexed data deleted, according to Mistral's own documentation. There's no automatic migration path. Administrators have to install MCP-based replacements themselves, and every user will need to reconnect their Google or Microsoft account from scratch once that's done.

The system being retired is, at least, well understood. An admin picks which Drive folders or SharePoint sites to expose, Mistral processes those files and parks the resulting index in its European data centers, and users then search against that prebuilt index while permission checks mirror what's set in Google or Microsoft. Scheduled syncs handle updates and deletions. Indexing itself can take anywhere from a few minutes to several hours depending on data volume, but that's finished before anyone starts searching.

What replaces it is murkier. MCP, as Mistral describes it, is simply a common interface letting models call tools and fetch data from external services — the same protocol showing up across the industry to connect AI products to outside APIs. Mistral quietly added Google Drive and SharePoint to a directory of more than 60 MCP integrations back in June, but never explained how those two connectors actually retrieve documents or who runs the servers behind them. MCP setups can mean live calls straight to the source API, a server-managed index, or some blend of both — and which one you get depends entirely on whoever operates that particular server. Mistral doesn't run those servers itself, so it can't guarantee behavior or make commitments about how customer data gets handled.

The permissions story gets murkier still. Today's Google Drive connector respects existing sharing rules — group and domain access, the fact that a link-shared file isn't automatically visible org-wide. The SharePoint connector relies on Microsoft Entra ID groups, which means older groups created purely inside SharePoint never get picked up. Mistral hasn't said whether the incoming MCP connectors will replicate any of this. OAuth can restrict what a server sees to a single connected user, but that's a different guarantee than filtering search results exactly the way the old index did — MCP wasn't built to enforce that kind of enterprise permission model, so the job falls to whatever retrieval layer each server implements.

Even the shutdown mechanics are unclear. Mistral says disabling a Knowledge Connector permanently deletes its indexed data, but the migration notice doesn't specify whether the August deadline triggers that deletion automatically or how long it takes, leaving admins guessing about whether they need to disconnect the old connectors manually beforehand. Because the same connector framework feeds Vibe Code and Mistral's workflow system too, whatever governance gaps exist here don't stay contained to one product — they ride along into coding and automation pipelines as well. Mistral's own advice to watch server output for prompt injection signs is a tidy admission of how unsettled this territory still is.