Mistral Agents API Release

Mistral AI has introduced the Agents API, a dedicated framework designed to transform traditional language models into active problem-solvers. By integrating built-in connectors, persistent memory, and orchestration capabilities, the Agents API allows developers to build enterprise-grade agents that can execute actions and maintain context across conversations.

Core Capabilities and Built-in Connectors

The Agents API provides a suite of pre-deployed tools known as built-in connectors, alongside support for the Model Context Protocol (MCP) to extend agent functionality.

Built-in Connectors

  • Code Execution: Agents can execute Python code within a secure sandboxed environment for mathematical calculations, data visualization, and scientific computing.
  • Image Generation: Powered by Black Forest Lab FLUX1.1 [pro] Ultra, this connector allows agents to create custom graphics and visual aids.
  • Document Library: This tool enables integrated Retrieval-Augmented Generation (RAG) by allowing agents to access user-uploaded documents stored in Mistral Cloud.
  • Web Search: This connector provides access to up-to-date information from the web and news sources. According to Mistral AI, adding web search significantly improves performance on the SimpleQA benchmark: Mistral Large improves from 23% to 75% accuracy, and Mistral Medium improves from 22.08% to 82.32%.

MCP Tool Integration

The Agents API SDK supports the Model Context Protocol (MCP), an open standard that enables agents to integrate with external systems such as databases, APIs, and dynamic user data. This allows for the extensible use of real-world context beyond the built-in connectors.

Stateful Conversation Management

Unlike traditional stateless completions, the Agents API implements a stateful conversation system that preserves context over time.

Conversation Structure

Developers can initiate interactions in two ways: by creating a conversation with a specific agent_id to use specialized capabilities, or by specifying the model and completion parameters directly for quick access to connectors.

Persistence and Branching

Because the API maintains a structured history of conversation entries, developers no longer need to manually monitor or pass back the entire conversation history. The system supports viewing past conversations, continuing existing threads, or initiating new conversation paths (branching) from any point in the history. Additionally, the API supports streaming outputs for real-time interaction.

Multi-Agent Orchestration and Handoffs

The Agents API enables the coordination of multiple specialized agents to solve complex problems through a process of dynamic orchestration.

Agentic Workflows

Developers can create multiple agents, each equipped with specific models and tools. By defining handoff rules, one agent can delegate tasks to another based on the conversation's requirements. For example, a finance agent may hand off a specific task to a web search agent or a calculator agent.

Collaborative Problem Solving

This handoff mechanism allows a single user request to trigger a chain of actions across multiple agents, allowing each to handle a specialized portion of the request for more efficient problem-solving.

Practical Applications

Mistral AI has provided several reference implementations (cookbooks) to demonstrate the API's versatility:

  • Coding Assistant: An agent that interacts with GitHub and oversees a developer agent powered by DevStral to automate software development tasks.
  • Linear Tickets Assistant: A multi-server MCP architecture that converts call transcripts into PRDs and subsequently into actionable Linear issues.
  • Financial Analyst: An agent that orchestrates multiple MCP servers to source financial metrics and archive insights.
  • Travel and Nutrition Assistants: Specialized agents for trip planning, booking management, and personalized dietary tracking.

Sources

Related

  • Dispatch
    Mistral AI Connectors release notes / what's new

    Mistral AI has released Connectors in Studio, enabling developers to integrate enterprise data via built-in and custom Model Context Protocol (MCP) connectors with programmatic API/SDK access.

  • Dispatch
    Mistral AI introduces granular admin controls, scoped API keys, and multi‑account connectors

    Mistral AI announced new connector features—including GA admin controls, scoped API keys, multi‑account connectors, a Debugger, and integrations in Workflows and Vibe Code—to improve security and reliability for enterprise AI workloads.

  • Project
    modelscope/ms-agent

    MS‑Agent is a lightweight Python framework for building autonomous LLM agents with tool‑calling, memory, multimodal input, and a skill‑retrieval system. It uses ModelScope’s Model Calling Protocol (MCP) for LLM access, provides a React/FastAPI web UI, and includes ready‑made projects for code generation, short video creation, and deep‑research workflows.

  • Dispatch
    Mistral Code Release Notes

    Mistral AI has launched Mistral Code, a vertically integrated AI coding assistant for enterprises that combines frontier models with an in-IDE assistant and flexible deployment options.

  • Dispatch
    Mistral AI Agentic Workflow for Product Development

    Mistral AI has introduced an agentic workflow, powered by Mistral Large 2, that automates the conversion of meeting transcripts into Product Requirements Documents (PRDs) and actionable engineering tickets.