LibreLink MCP Server

A local Model Context Protocol (MCP) server that provides Claude Desktop with secure access to your FreeStyle LibreLink continuous glucose monitoring (CGM) data.

🌟 Features

• Real-time glucose monitoring - Get current readings with trend arrows

• Historical data analysis - Retrieve glucose history over customizable periods

• Comprehensive analytics - Time-in-range, GMI, variability metrics

• mg/dL or mmol/L - Choose your unit; analytics stay accurate either way

• Pattern recognition - Dawn phenomenon, meal responses, stability analysis

• Privacy-first design - All data stays local on your machine

• Secure credential management - Local encrypted storage

• Cross-platform health integration - Works alongside other health MCP servers

Related MCP server: LibreLink MCP Server

📋 Prerequisites

• LibreLink Account: Active FreeStyle LibreLink account with glucose data

• Compatible Sensor: FreeStyle Libre 2 or 3 with data sharing enabled

• Node.js: Version 18.0.0 or higher

• Claude Desktop: For MCP integration

🚀 Quick Start

1. Installation

# Clone the repository
git clone https://github.com/donnycrash/librelink-mcp-http.git
cd librelink-mcp-http

# Install dependencies
npm install

# Build the project
npm run build

2. Configuration

# Configure your LibreLink credentials
npm run configure

You'll be prompted for:

• Email: Your LibreLink account email

• Password: Your LibreLink account password

• Region: US or EU (based on your location)

• Units: mg/dL (United States) or mmol/L (UK, Europe, South Africa, Australia)

• Target ranges: Glucose target ranges, entered in your chosen unit (default: 70-180 mg/dL, shown as 3.9-10.0 mmol/L)

3. Test Connection

# Test your LibreLink connection
node test-real-connection.js

4. Claude Desktop Integration

Add to your Claude Desktop configuration file:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "librelink": {
      "command": "node",
      "args": ["/path/to/librelink-mcp-server/dist/index.js"],
      "env": {
        "LIBRE_LINK_UP_VERSION": "4.16.0"
      }
    }
  }
}

The env block is required. Abbott rejects the client version pinned by the upstream API library with 403 / status 920 and a minimumVersion in the response body. Set LIBRE_LINK_UP_VERSION to at least that version. See Troubleshooting.

If you already have other MCP servers configured, merge the librelink key into the existing mcpServers object rather than replacing the file.

5. Restart Claude Desktop

Restart Claude Desktop to load the new MCP server.

🩸 Usage Examples

Once integrated with Claude Desktop, you can ask:

Basic Glucose Queries

• "What's my current glucose level?"

• "Show me my glucose readings from the past 6 hours"

• "What's my average glucose today?"

Analytics & Insights

• "Calculate my time in range for this week"

• "Analyze my glucose patterns and trends"

• "Do I have dawn phenomenon?"

• "How stable are my overnight glucose levels?"

Health Correlations

When combined with other health MCP servers:

• "How does my sleep quality affect my glucose control?"

• "Compare my glucose variability with my stress levels"

• "Show the impact of my supplements on glucose stability"

🛠 Available MCP Tools

📊 Sample Output

Samples below use the default mg/dL. With mmol/L configured, every glucose value converts and units reports mmol/L.

Current Glucose Reading

{
  "current_glucose": 105,
  "units": "mg/dL",
  "timestamp_local": "Jul 14, 2025, 11:19:24 PM",
  "timestamp_utc": "2025-07-14T21:19:24.000Z",
  "timezone": "Europe/London",
  "trend": "Flat",
  "status": "Normal",
  "color": "green"
}

Glucose Statistics

{
  "analysis_period_days": 7,
  "units": "mg/dL",
  "average_glucose": 93.46,
  "glucose_management_indicator_percent": 5.55,
  "time_in_range": {
    "target_range": "70-180 mg/dL",
    "in_range_percent": 100.0,
    "below_range_percent": 0.0,
    "above_range_percent": 0.0
  },
  "variability": {
    "standard_deviation": 7.52,
    "coefficient_of_variation_percent": 8.04
  }
}

GMI is an estimated A1C percentage and the variability figures are ratios, so they read the same in either unit. average_glucose and standard_deviation are concentrations and convert with the configured unit.

Trend Analysis

{
  "period": "daily",
  "units": "mg/dL",
  "patterns": [
    "Good postprandial glucose control",
    "Excellent overnight glucose stability"
  ],
  "dawn_phenomenon": false,
  "meal_response_average": 0,
  "overnight_stability": 2.08
}

Sensor Information

{
  "timezone": "Europe/London",
  "active_sensors": [
    {
      "device_id": "8389721a-5e5d-11ef-82f7-923037a589ee",
      "serial_number": "MH00XXXXXX",
      "device_type": "FreeStyle Libre 3",
      "state": "Active",
      "applied_local": "Sep 3, 2026, 11:05:28 AM",
      "applied_utc": "2026-09-03T09:05:28.000Z",
      "warmup_minutes": 60,
      "started_local": "Sep 3, 2026, 12:05:28 PM",
      "started_utc": "2026-09-03T10:05:28.000Z"
    }
  ],
  "sensor_count": 1
}

applied_* is when the sensor was attached; started_* adds the reported warmup period and is the figure the LibreLink app shows as the sensor start. Timestamps are given in both the host timezone and UTC.

🌐 Environment Variables

Environment values take precedence over config.json, so an injected secret is never shadowed by a stale file. Invalid LIBRELINK_REGION or LIBRELINK_UNITS values are ignored with a warning.

☁️ Remote Deployment (Northflank)

The default dist/index.js speaks MCP over stdio, which only works when a client launches it as a child process. For a hosted deployment the server also ships a Streamable HTTP entrypoint, dist/http.js, that listens on PORT, exposes POST /mcp behind bearer-token auth, and answers GET /health for platform health checks. The included Dockerfile builds that entrypoint.

Hosting this server moves your LibreLink password and glucose history onto a third-party platform. The local-only privacy posture described below does not apply to a remote deployment. Keep MCP_AUTH_TOKEN long and random and treat the URL as sensitive.

Northflank setup

1. Create a project, then add a Combined service linked to this repository and branch.

2. Build type Dockerfile, path /Dockerfile.

3. Under Networking, add port 3000, protocol HTTP, publicly exposed. Northflank issues a *.code.run URL with TLS.

4. Create a Secret group with LIBRELINK_EMAIL, LIBRELINK_PASSWORD, LIBRELINK_REGION, and MCP_AUTH_TOKEN (for example openssl rand -hex 32), and link it to the service. The Dockerfile defaults LIBRE_LINK_UP_VERSION to 4.16.0; add it to the group only if Abbott starts reporting a higher minimumVersion.

5. Set the health check to HTTP GET /health on port 3000.

6. Run a single replica. The glucose cache and the LibreLink login session live in process memory, and multiple replicas would each log in to Abbott separately.

Connecting a client

# Claude Code
claude mcp add --transport http librelink https://<your-service>.code.run/mcp \
  --header "Authorization: Bearer <MCP_AUTH_TOKEN>"

In Claude Desktop, add the same URL as a remote connector with the Authorization: Bearer <MCP_AUTH_TOKEN> header.

Running the HTTP server locally

npm run build
MCP_AUTH_TOKEN=dev-token npm run start:http
curl http://localhost:3000/health

Or with Docker:

docker build -t librelink-mcp .
docker run --rm -p 3000:3000 \
  -e MCP_AUTH_TOKEN=dev-token \
  -e LIBRELINK_EMAIL=you@example.com \
  -e LIBRELINK_PASSWORD=secret \
  librelink-mcp

🔧 Development

Running Tests

# Run all tests
npm test

# Test MCP protocol
npm run test:mcp

# Test analytics with mock data
npm run test:analytics

# Test environment-variable config overrides
npm run test:config

# Test the HTTP entrypoint (health, auth, tools/list)
npm run test:http

# Test with real LibreLink data (requires configuration)
node test-real-data.js

npm run test:mcp calls the configure_credentials tool with dummy values. It points LIBRELINK_MCP_CONFIG_DIR at a temporary directory it removes on teardown, so your real credentials in ~/.librelink-mcp/config.json are left untouched. Tests that exercise live data (test-real-connection.js, test-real-data.js) deliberately read the real configuration, and need LIBRE_LINK_UP_VERSION set in the environment.

Building

# Build TypeScript
npm run build

# Type checking
npm run typecheck

# Development mode
npm run dev

Project Structure

librelink-mcp-server/
├── src/
│   ├── index.ts              # Main MCP server
│   ├── librelink-client.ts   # LibreLink API wrapper
│   ├── glucose-analytics.ts  # Analytics and statistics
│   ├── config.ts             # Configuration management
│   ├── configure.ts          # CLI configuration tool
│   ├── units.ts              # mg/dL <-> mmol/L conversion
│   └── types.ts              # TypeScript definitions
├── config/
│   └── default.json          # Default configuration
├── test-*.js                 # Test suites
├── package.json
└── README.md

🔒 Security & Privacy

Data Privacy

• Local processing only - No data sent to external servers

• Your data stays on your machine - Complete privacy control

• No analytics or tracking - Zero telemetry

Credential Security

• Local storage - Credentials stored in ~/.librelink-mcp/config.json

• File permissions - Automatically set to user-only access (600)

• No cloud storage - Never uploaded or shared

Security Best Practices

# Verify file permissions
ls -la ~/.librelink-mcp/config.json
# Should show: -rw------- (user read/write only)

# Optional: Encrypt config directory
# (Implementation details in documentation)

⚠️ Important Notes

LibreLink API Usage

• This project uses an unofficial API through reverse engineering

• Not affiliated with Abbott or FreeStyle Libre

• Use at your own discretion and ensure compliance with LibreLink terms

• API may change - community maintained compatibility

Data Sharing Requirements

• Ensure your LibreLink app has data sharing enabled

• Your sensor must be active and transmitting data

• LibreLink account (not LibreLinkUp) credentials required

Sensor Compatibility

• ✅ FreeStyle Libre 2

• ✅ FreeStyle Libre 3

• ❓ FreeStyle Libre 1 (may work, not tested)

🐛 Troubleshooting

Common Issues

403 with {"status": 920, "data": {"minimumVersion": "..."}}

• Abbott is rejecting the client version, not your credentials — login succeeds and the failure appears on the next request

• Set LIBRE_LINK_UP_VERSION to at least the minimumVersion in the response

• If Abbott raises the minimum again, the symptom is identical; raise the value

"No connections found"

• The API authenticates against LibreLinkUp; verify your email and password sign in to the LibreLinkUp app itself

• Check that data sharing is enabled in your LibreLink app

• Ensure your sensor is active and connected

"Authentication failed"

• Double-check email and password

• Verify correct region (US/EU)

• Try logging into LibreLink app to confirm credentials

"Connection timeout"

• Check internet connection

• Verify LibreLink service status

• Try again after a few minutes

Getting Help

1. Run diagnostics:

   node diagnose-account.js

2. Check logs: Look for error messages in the console output

3. Test connection:

   node test-real-connection.js

4. Open an issue: Include diagnostic output and error messages

🤝 Contributing

We welcome contributions! Please:

1. Fork the repository

2. Create a feature branch: git checkout -b feature/amazing-feature

3. Make your changes with tests

4. Follow the existing code style

5. Submit a pull request

Development Guidelines

• TypeScript required - Maintain type safety

• Test coverage - Add tests for new features

• Documentation - Update README for new functionality

• Security first - Never commit credentials or sensitive data

📜 License

This project is licensed under the MIT License - see the LICENSE file for details.

🙏 Acknowledgments

• libre-link-unofficial-api - Community-maintained LibreLink API client

• MCP Protocol - Anthropic's Model Context Protocol

• FreeStyle Libre Community - Inspiration and reverse engineering efforts

• Open Source Diabetes Projects - Nightscout, OpenAPS, and others

⭐ Support

If this project helps you manage your diabetes with AI assistance, please:

• ⭐ Star the repository

• 🐛 Report issues you encounter

• 💡 Suggest improvements

• 🤝 Contribute to the project

Disclaimer: This is an unofficial project not affiliated with Abbott or FreeStyle Libre. Use responsibly and in compliance with applicable terms of service. Always consult healthcare professionals for medical decisions.