Mistral Overhauls Connectors With Admin Controls, Debugging, and Multi-Account Support

Mistral has rolled out a suite of connector upgrades designed to solve what it sees as the hardest part of deploying AI agents in production: reliably connecting them to enterprise tools without breaking security or compliance.

The headline addition is enriched admin controls, now generally available. Administrators can set connector access per workspace and toggle individual tools on or off across an entire organization or a single team. A finance workspace, for instance, could block all web access while engineering gets developer tools — same directory, different rules. Tool-level controls let admins nuke specific actions like delete_file while keeping the rest of a connector live.

Identity gets sharper too. New API keys with connector scopes let teams limit what an automated job can reach, and when paired with service accounts, work runs as a defined identity — never impersonating whoever wrote the code. Multi-account connectors let a single connector hold multiple logins, each stored and refreshed independently, so an agent can switch between personal and work accounts per task without standing up a second connector.

For developers tired of guessing why a connection broke, the new Connectors Debugger walks an MCP connection through 11 steps — from reaching the server to opening the session — and logs each one. A failing OAuth token exchange, for example, surfaces at the exact step it dies. The tool is in public preview alongside Connectors in Workflows, which lets long-running or scheduled jobs declare their connector dependencies upfront so tokens don't expire mid-run.

The broader signal here is that Mistral is treating agent connectivity as an infrastructure problem, not a demo trick. The company now lists more than 60 integrations and offers custom MCP connectors for everything else. Enterprise controls, credential scopes, and Vibe Code integration are live now in Studio.