brown-tech-info/PubMed-MCP-Azure-Function

PubMed MCP Azure Function

This project is a Python-based MCP (Model Context Protocol) server for PubMed, hosted as an Azure Function and integrated with Azure AI Foundry.

Features

• Search PubMed articles and return titles and abstracts via HTTP requests.
• Uses PubMed Entrez API with API key and email for authentication.
• Deployed on Azure Functions (Linux, Python 3.11).
• Anonymous public endpoint (no function key required).
• OpenAPI 3.0.1 specification for agent/tool integration.

Setup

Prerequisites

• Python 3.11
• Azure Functions Core Tools
• Azure CLI
• An Azure subscription
• PubMed API key and registered email

Local Development

1. Clone the repository.
2. Install dependencies:

   pip install -r requirements.txt
   

3. Add your PubMed API key and email to local.settings.json:

   {
     "IsEncrypted": false,
     "Values": {
       "AzureWebJobsStorage": "UseDevelopmentStorage=true",
       "FUNCTIONS_WORKER_RUNTIME": "python",
       "PUBMED_API_KEY": "<your_pubmed_api_key>",
       "PUBMED_EMAIL": "<your_email>"
     }
   }
   

4. Start the function locally:

   func start
   

5. Test locally:

   curl "http://localhost:7071/api/pubmed_mcp?query=cancer"
   

Azure Deployment

1. Log in to Azure:

   az login
   

2. Create a resource group:

   az group create --name <ResourceGroup> --location <Region>
   

3. Create a storage account:

   az storage account create --name <StorageAccount> --location <Region> --resource-group <ResourceGroup> --sku Standard_LRS
   

4. Create a Function App (Linux, Python 3.11):

   az functionapp create --resource-group <ResourceGroup> --consumption-plan-location <Region> --runtime python --runtime-version 3.11 --functions-version 4 --name <FunctionAppName> --storage-account <StorageAccount> --os-type Linux
   

5. Publish your code:

   func azure functionapp publish <FunctionAppName>
   

6. Set environment variables in Azure:

   az functionapp config appsettings set --name <FunctionAppName> --resource-group <ResourceGroup> --settings PUBMED_API_KEY="<your_pubmed_api_key>" PUBMED_EMAIL="<your_email>"
   

7. Test the deployed function (no key required):

   curl "https://<FunctionAppName>.azurewebsites.net/api/pubmed_mcp?query=cancer"
   

Agent & Tool Integration

• The function is compatible with Azure AI Foundry agents and other OpenAPI-based tools.
• The OpenAPI spec (openapi.json) describes the endpoint for agent integration.
• Example agent prompt:

  Find recent PubMed articles about cancer and return their titles and abstracts.

Security

• Do not commit your real API keys or sensitive information to source control.
• Use environment variables for secrets.

Logging

• The function logs requests, errors, and PubMed API responses to the Azure log stream.

Customization

• You can extend the MCP server to support more PubMed features or custom endpoints.

License

MIT