The Mistral Agents API is a hosted service for building agents on Mistral's language models, extending the Chat Completion API with persistent memory, built-in connectors, and agentic orchestration. It targets use cases where an agent needs to maintain context across a conversation and coordinate multiple actions rather than produce a single response.
Agents can be equipped with built-in connectors, including a sandboxed environment for executing Python code, web search, and image generation, as well as Model Context Protocol (MCP) tools and multi-server MCP setups. Documented example workflows include a coding assistant that interacts with a code-hosting platform, a task-coordination assistant that turns call transcripts into product-requirement documents and tracked issues, and a financial-advisory agent that orchestrates several MCP servers to source and archive data.
Features
- Persistent memory: conversations retain context across multiple turns and sessions
- Built-in connectors: code execution, web search, and image generation are available as callable tools
- Sandboxed code execution: agents run Python in a secure sandboxed environment for calculations, data analysis, and plotting
- MCP tool support: agents call tools exposed by one or more Model Context Protocol servers
- Agentic orchestration: an agent coordinates multiple actions and tool calls to complete a task
