Foundry Medical Research Agents

Azure AI Foundry的医学研究MCP代理

该项目提供了一系列 MCP（模型上下文协议）服务器 用于医学和生物医学研究，部署到Azure容器应用程序，并用作Azure AI Foundry代理中的工具。

🏥 可用代理

建筑

┌─────────────────────────────────────────────────────────────┐
│                    Azure AI Foundry                         │
│  ┌─────────────────────────────────────────────────────┐   │
│  │           Declarative Agents (YAML)                  │   │
│  │  - PubMed Research Assistant                        │   │
│  │  - Clinical Trials Assistant                        │   │
│  │  - FDA Drug/Device Assistant                        │   │
│  │  - Protein Research Assistant                       │   │
│  └──────────────────────┬──────────────────────────────┘   │
└─────────────────────────┼───────────────────────────────────┘
                          │ HTTP (MCP JSON-RPC)
                          ▼
┌─────────────────────────────────────────────────────────────┐
│              Azure Container Apps                           │
│  ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ ┌────────┐│
│  │ PubMed MCP  │ │ClinicalTrial│ │ OpenFDA MCP │ │UniProt ││
│  │   Server    │ │  MCP Server │ │   Server    │ │  MCP   ││
│  └──────┬──────┘ └──────┬──────┘ └──────┬──────┘ └───┬────┘│
└─────────┼───────────────┼───────────────┼────────────┼─────┘
          │               │               │            │
          ▼               ▼               ▼            ▼
    ┌──────────┐   ┌────────────┐   ┌──────────┐  ┌─────────┐
    │  NCBI    │   │ClinicalTri-│   │ OpenFDA  │  │ UniProt │
    │ PubMed   │   │ als.gov    │   │   API    │  │REST API │
    │   API    │   │    API     │   │          │  │         │
    └──────────┘   └────────────┘   └──────────┘  └─────────┘

快速开始

先决条件

1. Azure帐户 与：

- Azure AI Foundry项目 - 用于托管的Azure容器应用程序（或类似应用程序） - Azure CLI已安装并登录(az login)

1. API密钥 （根据需要）：

- NCBI API PubMed密钥（免费）- NCBI帐户设置 - ClinicalTrials.gov、OpenFDA和UniProt API是免费的，不需要密钥

______________________________________________________________________

📚 PubMed MCP服务器

说明： 从NCBI PubMed数据库中搜索和检索医学文献。访问摘要、引文、作者信息和研究趋势。

GitHub来源： 基督教/基督教联合会

MCP端点： https://pubmed-mcp-server.orangemeadow-5c32aca2.eastus.azurecontainerapps.io/mcp

工具（12个工具）

代理说明

You are a specialized medical and scientific literature research assistant with access to PubMed.

Your capabilities include:
- Advanced PubMed literature search with complex filters
- Retrieving detailed article information with abstracts
- Searching by specific authors and finding co-authors
- Discovering related articles
- Searching and exploring MeSH terms
- Analyzing journals and research trends
- Exporting citations in multiple formats

When users ask for literature searches:
1. Use appropriate filters (date ranges, article types, etc.)
2. Provide concise summaries of key findings
3. Include relevant PMIDs for reference
4. Offer to export citations if needed

Always be thorough, accurate, and cite your sources properly.

______________________________________________________________________

🏥 ClinicalTrials.gov MCP服务器

说明： 从ClinicalTrials.gov搜索和分析临床试验。按条件、干预、地点、赞助商查找试验，并分析资格标准。

GitHub来源： JackKenney/临床试验医学博士 （参考）

MCP端点： https://clinicaltrials-mcp-server.orangemeadow-5c32aca2.eastus.azurecontainerapps.io/mcp

工具（6个工具）

代理说明

You are a clinical trials research assistant. Use the [clinicaltrials-tools] to help users find information about clinical trials. You MUST NEVER answer from your own knowledge. If you cannot find the answer using the available tools, respond with "I don't know".

When users ask about clinical trials:
1. Use [search_clinical_trials] to find trials by condition, intervention, or location
2. Use [get_study_details] when users have a specific NCT ID
3. Use [get_eligibility_criteria] when users want to know who can participate
4. Use [get_study_locations] to find trial sites near a location
5. Use [search_by_sponsor] to find trials by pharmaceutical companies or institutions
6. Use [get_trial_statistics] for enrollment and study design information

Always cite the NCT ID and source as ClinicalTrials.gov data.

______________________________________________________________________

💊 OpenFDA MCP服务器

说明： 访问FDA药品和医疗器械数据，包括不良事件、召回、标签和监管许可。

GitHub来源： williamlu2000/openFDA-mcp （参考）

MCP端点： https://openfda-mcp-server.orangemeadow-5c32aca2.eastus.azurecontainerapps.io/mcp

工具（10个工具）

药物工具（6）：

设备工具（4）：

代理说明

You are an FDA drug and medical device research assistant. Use the [openfda-tools] to help users find information about FDA-regulated products. You MUST NEVER answer from your own knowledge. If you cannot find the answer using the available tools, respond with "I don't know".

When users ask about FDA-regulated products:
1. Use [search_drug_adverse_events] for drug side effects and safety reports
2. Use [search_drug_labels] for prescribing information, warnings, and dosage
3. Use [search_drug_ndc] to look up drugs by NDC or find generic equivalents
4. Use [search_drug_recalls] for drug recalls by class or manufacturer
5. Use [search_drug_shortages] for current drug supply issues
6. Use [search_device_510k] for FDA-cleared medical devices
7. Use [search_device_classifications] for device regulatory classifications
8. Use [search_device_adverse_events] for device malfunction or injury reports
9. Use [search_device_recalls] for medical device recalls

Always cite the source as FDA/openFDA data.

______________________________________________________________________

🧬 UniProt MCP服务器

说明： 从UniProt访问蛋白质和蛋白质组学数据，包括序列、结构、结构域、变体、相互作用和基因本体注释。

GitHub来源： BioContext/UniProt MCP （参考）

MCP端点： https://uniprot-mcp-server.orangemeadow-5c32aca2.eastus.azurecontainerapps.io/mcp

工具（8个工具）

代理说明

You are a protein and proteomics research assistant. Use the [uniprot-tools] to help users find information about proteins. You MUST NEVER answer from your own knowledge. If you cannot find the answer using the available tools, respond with "I don't know".

When users ask about proteins:
1. Use [get_protein_by_accession] for specific UniProt IDs (e.g., P01308 for insulin)
2. Use [search_proteins] to find proteins by gene name, name, or organism
3. Use [get_protein_sequence] for amino acid sequences in FASTA format
4. Use [get_protein_features] for domains, variants, and modifications
5. Use [get_protein_structure] for PDB structures and AlphaFold predictions
6. Use [get_protein_interactions] for binding partners and complexes
7. Use [get_protein_go_terms] for function and cellular localization
8. Use [search_by_sequence] to identify unknown sequences

Common accessions: P01308 (insulin), P38398 (BRCA1), P04637 (p53), P00533 (EGFR)

Always cite the source as UniProt/Swiss-Prot data.

______________________________________________________________________

部署

部署所有服务器

每个MCP服务器都可以部署到Azure容器应用程序：

# Create resource group and environment (one-time)
az group create --name pubmed-mcp-rg --location eastus
az containerapp env create --name pubmed-mcp-env --resource-group pubmed-mcp-rg --location eastus

# Deploy PubMed MCP Server
cd pubmed-mcp
az containerapp up --name pubmed-mcp-server --resource-group pubmed-mcp-rg --environment pubmed-mcp-env --source . --env-vars PUBMED_API_KEY=your_key PUBMED_EMAIL=your_email --target-port 8000 --ingress external

# Deploy ClinicalTrials MCP Server
cd ../clinicaltrials-mcp
az containerapp up --name clinicaltrials-mcp-server --resource-group pubmed-mcp-rg --environment pubmed-mcp-env --source . --target-port 8000 --ingress external

# Deploy OpenFDA MCP Server
cd ../openfda-mcp
az containerapp up --name openfda-mcp-server --resource-group pubmed-mcp-rg --environment pubmed-mcp-env --source . --target-port 8000 --ingress external

# Deploy UniProt MCP Server
cd ../uniprot-mcp
az containerapp up --name uniprot-mcp-server --resource-group pubmed-mcp-rg --environment pubmed-mcp-env --source . --target-port 8000 --ingress external

连接到Azure AI Foundry

1. 首选 ai.ai.com 或使用VS Code Foundry扩展
2. 创建新代理
3. 添加一个 MCP服务器工具 与：

- 服务器URL: https://.orangemeadow-5c32aca2.eastus.azurecontainerapps.io/mcp - 认证：未经身份验证

1. 添加上述章节中的代理说明
2. 部署和测试

______________________________________________________________________

项目结构

foundry-pubmed-agent/
├── pubmed-mcp/                    # PubMed MCP Server
│   ├── src/
│   │   ├── http_server.py         # HTTP/MCP wrapper
│   │   ├── server.py              # Core MCP server
│   │   └── pubmed_client.py       # PubMed API client
│   ├── Dockerfile
│   └── requirements.txt
├── clinicaltrials-mcp/            # ClinicalTrials.gov MCP Server
│   ├── http_server.py             # HTTP/MCP wrapper
│   ├── Dockerfile
│   └── requirements.txt
├── openfda-mcp/                   # OpenFDA MCP Server
│   ├── http_server.py             # HTTP/MCP wrapper
│   ├── Dockerfile
│   └── requirements.txt
├── uniprot-mcp/                   # UniProt MCP Server
│   ├── http_server.py             # HTTP/MCP wrapper
│   ├── Dockerfile
│   └── requirements.txt
├── agents/                        # Agent definitions
├── config/                        # Configuration files
└── README.md

本地开发

在本地测试任何服务器

cd   # e.g., pubmed-mcp, openfda-mcp, etc.

# Create and activate virtual environment
python -m venv .venv
.venv\Scripts\Activate.ps1

# Install dependencies
pip install -r requirements.txt

# Run the server
python http_server.py  # or python -m src.http_server for pubmed-mcp

服务器将在以下时间可用 http://localhost:8000.测试终点：

# Health check
curl http://localhost:8000/health

# List tools
curl http://localhost:8000/tools

# Call a tool
curl -X POST http://localhost:8000/tools/call -H "Content-Type: application/json" -d '{"name": "tool_name", "arguments": {...}}'

# MCP endpoint (for Foundry)
curl -X POST http://localhost:8000/mcp -H "Content-Type: application/json" -d '{"jsonrpc":"2.0","method":"tools/list","params":{},"id":1}'

故障排除

代理无法连接到MCP服务器

• 验证服务器URL是否以结尾 /mcp
• 检查容器应用程序是否启用了外部入口
• 测试运行状况端点： curl https:///health

工具返回错误

• 检查特定API的费率限制
• 验证查询参数是否正确
• 检查容器日志： az containerapp logs show -n -g pubmed-mcp-rg

API特定问题

• PubMed：需要API密钥以获得更高的速率限制（3/秒与1/秒）
• ClinicalTrials.gov：不需要身份验证，但复杂的查询可能会超时
• OpenFDA：无需身份验证，每分钟限制240个请求
• UniProt：不需要身份验证，但大型查询可能需要时间

资源

• Azure AI Foundry MCP集成
• PubMed MCP服务器
• 临床试验MCP
• OpenFDA MCP
• UniProt MCP
• NCBI电子公用事业
• ClinicalTrials.gov API
• OpenFDA API
• UniProt REST API

许可证

MIT许可证