Mistral AI · 2026-04-15 · notable
Mistral Connectors — MCP and Built-in Data Sources via API and SDK
Mistral opens all built-in connectors (GitHub, Gmail, web search) and custom MCP servers to its API and SDK, adding direct tool calling and human-in-the-loop approval flows. Enterprise integrations now programmable across all Mistral models.
Mistral's built-in connectors and custom MCP servers are now accessible programmatically through its API and SDK.
What is it?
Mistral AI's Connectors feature — previously limited to the Studio UI — is now available via the Conversation API, Completions API, and Agent SDK. Developers can integrate enterprise data sources (GitHub, Gmail, web search, and custom MCP servers) directly into any model or agent call, with connectors registered once and shared across teams and platforms.
How does it work?
Connectors are packaged using the MCP protocol and registered centrally in Studio, making them discoverable across LeChat, AI Studio, and Mistral Vibe. Direct tool calling lets developers invoke a specific connector deterministically without model selection. A `requires_confirmation` flag pauses execution pending human review before a tool runs. Connectors are created, modified, listed, and deleted programmatically via the API.
Why does it matter?
Teams previously rebuilt OAuth and API integrations for each project. A single registered connector is now reused across all model calls with consistent access controls and audit trails. The human-in-the-loop approval mechanism makes it practical to expose sensitive tools — write access to email or code repos — in production agents.
Who is it for?
Enterprise developers building agentic workflows that read or write to internal systems like code repos, email, or CRMs.
Try it
https://docs.mistral.ai/capabilities/agents/