Mistral unveils Mistral Agents API

Artificial intelligence startup Mistral AI Tuesday announced the Mistral Agents API, a complement to its Chat Completion API that “simplifies implementing agentic use cases,” the company said.

The Agents API combines Mistral’s large language models (LLMs) with built-in connectors for code execution, web search, image generation, and Model Context Protocol (MCP) tools, persistent memory across conversations, and agentic orchestration functionality to address an LLM’s limitations.

Mistral Agents API connectors

The Mistral Agents API features three connectors. The code execution connector lets developers build agents that execute Python code in a secure sandbox. The image generation connector allows agents to create images; it’s powered by BlackForestLabs FLUX1.1 [pro] Ultra. The document library connector lets agents access documents in Mistral Cloud; it also powers the integrated retrieval-augmented generation (RAG) functionality.

The web search connector allows developers to combine Mistral models with fresh information sourced online. The company said that agents with web search capabilities “show a significant improvement in performance.” It said that in the SimpleQA benchmark, Mistral Large and Mistral Medium with web search achieved scores of 75% and 82.32%, respectively, compared to 23% and 22.08% without web search.

The Agents API provides conversation management that gives developers the ability to maintain conversation state, continuing existing conversations or initiating new ones as required. It supports streaming outputs to allow for real-time updates and interactions.

However, Mistral noted, “The true power of our Agents API lies in its ability to orchestrate multiple agents to solve complex problems. Through dynamic orchestration, agents can be added or removed from a conversation as needed—each one contributing its unique capabilities to tackle different parts of a problem.”

Mistral closes the gap

However, Brian Jackson, principal research director at Info-Tech Research Group, doesn’t view the API as a groundbreaking product.

“Mistral’s new Agents API is best read as a parity play and not a market-shifting release,” Jackson said. “It puts them in line with OpenAI’s Assistants, Google’s Gemini Agents, and Anthropic’s Claude tools, which offer persistent memory, tool-calling hooks, and structured  output, all of which are now table stakes in the space. It closes a feature gap for those already betting on Mistral’s models, but it doesn’t introduce functionality that the market hasn’t already seen.

This interesting part of the announcement, Jackson noted, was the MCP support.

“What does matter is the standard’s direction. It seems promising that all the major LLM players are integrating around Model Context Protocol,” he said. “It means that AI agents are going to be highly interoperable and that enterprises will only have to code their interface between legacy systems and AI agents once instead of several times. That implies that we’ll have an agentic AI ecosystem that’s more flexible and is less prone to vendor lock-in.”