📄 中文摘要
当前的自主科学研究（ASR）系统虽然利用了大型语言模型（LLMs）和智能架构，但仍受限于固定的工作流程和工具集，无法适应不断变化的任务和环境。Mimosa框架的提出旨在解决这一问题，通过自动合成特定任务的多智能体工作流程，并通过实验反馈进行迭代优化。Mimosa利用模型上下文协议（MCP）进行动态工具发现，通过元调度器生成工作流程拓扑，执行子任务的代码生成代理调用可用工具和科学软件库，并通过基于LLM的评估者对执行结果进行评分，从而推动工作流程的改进。
Mimosa框架：迈向演化的多智能体系统以支持科学研究
📄 English Summary
Mimosa Framework: Toward Evolving Multi-Agent Systems for Scientific Research
Current Autonomous Scientific Research (ASR) systems, despite leveraging large language models (LLMs) and agentic architectures, are constrained by fixed workflows and toolsets that hinder their adaptability to evolving tasks and environments. The Mimosa framework addresses this challenge by automatically synthesizing task-specific multi-agent workflows and iteratively refining them based on experimental feedback. It employs the Model Context Protocol (MCP) for dynamic tool discovery, generates workflow topologies through a meta-orchestrator, executes subtasks via code-generating agents that invoke available tools and scientific software libraries, and evaluates executions with an LLM-based judge whose feedback drives workflow improvements.