Mistral AI has announced a raft of upgrades to its Connectors platform, aimed at giving enterprises tighter governance over how AI agents access third-party tools and data.
The headline additions include enriched admin controls for setting connector access per workspace, API keys with connector scopes to stop automated workloads impersonating users, multi-account connectors letting one login point hold several accounts, a debugging tool for diagnosing broken MCP connections, connector support inside Vibe Code, and connector support inside Workflows for long-running tasks. Most of these are now generally available, though the debugger and the Workflows integration remain in public preview.
The company's pitch centers on a common enterprise frustration: it's straightforward to wire up an agent to enterprise data for a demo, but keeping that connection stable and properly governed once it's running in production is a much harder problem. The new tool-level controls address part of that by letting admins approve or block individual actions within a connector, rather than granting or denying access to the whole thing, so a write or delete action can be switched off while the rest of the connector keeps working.
Mistral says its connector directory now covers more than 60 integrations, spanning categories from knowledge and data tools to developer platforms, with custom MCP connectors available to fill any remaining gaps.
