Mimosa Framework: Toward Evolving Multi-Agent Systems for Scientific Research
Summary: arXiv:2603.28986v1 Announce Type: new
Abstract: Current Autonomous Scientific Research (ASR) systems, despite leveraging large language models (LLMs) and agentic architectures, remain constrained by fixed workflows and toolsets that prevent adaptation to evolving tasks and environments. We introduce Mimosa, an evolving multi-agent framework that automatically synthesizes task-specific multi-agent workflows and iteratively refines them through experimental feedback.
Mimosa leverages the Model Context Protocol (MCP) for dynamic tool discovery, generates workflow topologies via a meta-orchestrator, executes subtasks through code-generating agents that invoke available tools and scientific software libraries, and scores executions with an LLM-based judge whose feedback drives workflow refinement. On ScienceAgentBench, Mimosa achieves a success rate of 43.1% with DeepSeek-V3.2, surpassing both single-agent baselines and static multi-agent configurations.
Our results further reveal that models respond heterogeneously to multi-agent decomposition and iterative learning, indicating that the benefits of workflow evolution depend on the capabilities of the underlying execution model. Beyond these benchmarks, Mimosa’s modular architecture and tool-agnostic design make it readily extensible, and its fully logged execution traces and archived workflows support auditability by preserving every analytical step for inspection and potential replication.
Key Features of the Mimosa Framework
- Dynamic Tool Discovery: Mimosa utilizes the Model Context Protocol (MCP) to identify and integrate tools on-the-fly, ensuring that the most relevant resources are always available for task execution.
- Meta-Orchestrator: The framework features a meta-orchestrator that generates optimal workflow topologies, adapting to the specific needs of each scientific inquiry.
- Code-Generating Agents: Specialized agents are designed to execute subtasks by generating code that interacts with various scientific software libraries, enabling seamless integration and execution of complex tasks.
- LLM-Based Feedback Loop: A dedicated judge, powered by a large language model, evaluates the performance of workflows and provides feedback that is critical for iterative refinement.
- Extensibility and Auditability: The modular design of Mimosa allows for easy extension and integration of new tools, while comprehensive logging of execution traces ensures that all steps can be audited and replicated.
Potential Applications in Scientific Research
With its innovative approach to multi-agent systems, Mimosa has the potential to revolutionize various domains of scientific research, including but not limited to:
- Data Analysis: Automating the processing and analysis of large datasets across disciplines.
- Experimental Design: Assisting researchers in designing experiments tailored to specific hypotheses and conditions.
- Model Development: Facilitating the creation and validation of predictive models in fields such as biology, chemistry, and physics.
- Interdisciplinary Research: Enabling collaboration across different scientific fields by providing adaptable workflows.
Released as a fully open-source platform, Mimosa aims to provide an open foundation for community-driven Autonomous Scientific Research, paving the way for enhanced collaboration and innovation in scientific endeavors.