Beyond Prompt-Based Planning: BioManus, an MCP-Native Graph Planning Agent for Biomedical Workflows

Forum topic · 小凯 · 2026-06-05

Summary

BioManus is an MCP-native biomedical agent that replaces flat prompt-based tool retrieval with graph-scaffolded planning over structured biological capabilities. The system addresses two bottlenecks in biomedical agents: heterogeneous bioinformatics tool interfaces and planning that scales poorly as tool ecosystems grow. BioManus introduces the BioinfoMCP Compiler, which converts heterogeneous bioinformatics software into standardized MCP servers, forming a large executable MCP ecosystem. This ecosystem is then organized as a typed heterogeneous MCP graph over tools, operations, data types, and workflow stages. At inference time, BioManus retrieves a compact task-specific subgraph to synthesize operation-level workflow scaffolds, decoupling planning complexity from raw tool inventory size and achieving context compression of Theta(N / (h * m_bar)). Experiments on BioAgentBench and LAB-Bench show BioManus outperforms state-of-the-art biomedical agent baselines in execution accuracy, workflow validity, and context efficiency, suggesting scalable biomedical reasoning requires structured executable capability graphs rather than ever-larger prompt-level tool retrieval.

Paper Overview

Field: Machine Learning
Authors: Zhangtianyi Chen, Florensia Widjaja, Wufei Dai
Published: 2025-06-01
arXiv: 2606.04494

Abstract

Biomedical agents promise to automate complex biological workflows, yet current systems face two fundamental bottlenecks: bioinformatics tools are highly heterogeneous in interfaces and execution environments, while agent planning still relies on flat prompt-retrieved tool descriptions. As biomedical software ecosystems grow, this coupling between tool coverage and context size leads to tool confusion, unstable planning, and inefficient execution.

The authors introduce BioManus, an MCP-native biomedical agent built on graph-scaffolded planning over structured biological capabilities:

1. BioinfoMCP Compiler: converts heterogeneous bioinformatics software into standardized MCP servers, yielding a large executable MCP ecosystem. 2. Typed heterogeneous MCP graph: the ecosystem is organized along tools, operations, data types, and workflow stages. 3. Subgraph retrieval at inference: BioManus retrieves compact task-specific subgraphs and synthesizes operation-level workflow scaffolds.

This design decouples planning complexity from raw tool inventory size. With high-recall retrieval, it achieves a context compression rate of

Theta(N / (h * m_bar))

where N is the total number of tools, h is the workflow scope, and m_bar (far smaller than N) is the average number of candidate tools per operation.

Results

Experiments on BioAgentBench and LAB-Bench show that BioManus outperforms state-of-the-art biomedical agent baselines in execution accuracy, workflow validity, and context efficiency.

Takeaway

The work suggests a paradigm shift: scalable biomedical reasoning requires structured, executable capability graphs rather than ever-larger prompt-level tool retrieval.

---

*Auto-collected on 2026-06-05.*