🧬 SciAgent (mcp-server-bioscience)

English | 简体中文

SciAgent (BioScience MCP Server) is a high-performance, security-focused Model Context Protocol (MCP) server that empowers AI Agents with cutting-edge biological, medical, and scientific capabilities.

By connecting your AI Assistant (Claude Desktop, Cursor, Claude Code, Aider, etc.) to SciAgent, you eliminate AI hallucinations regarding scientific facts and instantly grant it the ability to interact with real-world, authoritative scientific databases.

🌟 Core Capabilities

• 📚 Academic Literature Mining: Direct access to PubMed and Europe PMC for the latest peer-reviewed papers.

• 🏥 Clinical Trials Data: Query global clinical trial data from ClinicalTrials.gov by disease, drug, phase, or status.

• 🧬 Genomic Data Analysis: Query genetic variants, gene coordinates, and metadata from Ensembl and dbSNP.

• 🧪 Drug Discovery: Fetch molecule properties and drug targets from ChEMBL.

• 🔬 Structural Biology: Retrieve and analyze 3D protein structures directly from AlphaFold and UniProt.

Related MCP server: grant-mcp

🏗️ Architecture

graph LR
    A[AI Agent<br/>Claude/Cursor] <-->|Model Context Protocol| B((SciAgent<br/>MCP Server))
    B <--> C[(PubMed)]
    B <--> D[(AlphaFold)]
    B <--> E[(ChEMBL)]
    B <--> F[(Ensembl)]
    style B fill:#2b5c8f,stroke:#fff,stroke-width:2px,color:#fff

🛡️ Built for AI Safety (Codex Security Ready)

When AI processes third-party biological data (such as parsing massive PDB files or reading unverified scientific texts), it is highly susceptible to Prompt Injection and Context Overflow.

SciAgent is engineered with a security-first approach:

• Sanitizer Middleware: Strictly validates API responses.

• Context Management: Truncates and summarizes massive biological payloads before handing data back to the LLM.

🚀 Quick Start

1. Installation

This package requires uv or pip. We highly recommend using uv for maximum speed.

# Clone the repository
git clone https://github.com/wangshuaifbj-collab/mcp-server-bioscience.git
cd mcp-server-bioscience

# Install
uv pip install -e .

(Ensure that the mcp-bioscience executable is added to your system's PATH after installation).

2. Connect to Your AI

For Claude Desktop

Add the following configuration to your claude_desktop_config.json:

{
  "mcpServers": {
    "bioscience": {
      "command": "mcp-bioscience"
    }
  }
}

For Cursor

1. Go to Settings -> Features -> MCP

2. Click + Add New MCP Server

3. Choose command type

4. Set Name: bioscience

5. Set Command: mcp-bioscience

For Claude Code

Run the following command in your terminal to easily add the server:

claude mcp add bioscience -- mcp-bioscience

💬 Example Use Case

You: "Can you find the latest papers on the mechanism of Imatinib and show me its primary protein target structure?"

Claude (powered by SciAgent):

1. Calls search_literature (PubMed) to find the latest Imatinib papers.

2. Calls get_gene_info to identify BCR-ABL1.

3. Calls get_protein_structure (AlphaFold) to return the structural confidence metrics of the target.

4. Provides you with a comprehensive, fully-cited response without hallucinations!

🤝 Contributing

We welcome contributions! Whether it's adding a new database (like KEGG or GEO) or improving the existing parsers, please feel free to open a Pull Request. This project aims to be the standard bioinformatics toolkit for the global Agent ecosystem.