---
title: Wellness CGM MCP connector
description: The Levels-killer pattern as a free local-first MCP — Dexcom + meal-glucose correlation.
canonical: https://wellness.delx.ai/connectors/cgm
content_signal: ai-train=no, search=yes, ai-input=yes
---

# Wellness CGM MCP Connector

Pair cgm_meal_response with wellness-nourish meal logs to get 'this meal hit a 72 mg/dL spike at 45min — try protein-first next time'. Levels charges $199/mo for this; here it's an MCP. Sandbox mode lets any agent demo without setup.

## Install

```bash
npx -y wellness-cgm-mcp doctor
npx -y wellness-cgm-mcp doctor
```

## Agent client config

```json
{
  "mcpServers": {
    "wellness-cgm": {
      "command": "npx",
      "args": ["-y", "wellness-cgm-mcp"],
      "env": {
        "DEXCOM_ENV": "sandbox",
        "DEXCOM_CLIENT_ID": "<your_id>",
        "DEXCOM_CLIENT_SECRET": "<your_secret>",
        "DEXCOM_REDIRECT_URI": "http://localhost:3012/callback",
        "DEXCOM_ACCESS_TOKEN": "<set_after_authorize>"
      }
    }
  }
}
```

## Signals

- Glucose
- TIR (70-180)
- TIR (70-140)
- GMI (estimated A1C)
- CV
- Meal response band

## Data availability

- Live EGV (5-minute glucose): available. Dexcom v3 API; mock fallback otherwise.
- Time-in-range (ADA 70-180): available. Computed from EGVs.
- Time-in-range (Levels 70-140): available. Metabolic-health profile for non-DM users.
- GMI (estimated A1C): available. Bergenstal 2018 formula: 3.31 + 0.02392 × mean(mg/dL).
- CV (variability): available. stdev / mean × 100.
- Meal response (peak + band): available. Pair with wellness-nourish meal logs.
- Refresh-token rotation: opt_in. Lands in v0.2.
- FreeStyle Libre (LibreLink Up): not_available. Roadmapped for v0.2.

## MCP tools

### Profile and onboarding

- `cgm_profile_get`
- `cgm_profile_update`
- `cgm_onboarding`

### Start here

- `cgm_quickstart`
- `cgm_demo`
- `cgm_glucose_now`

### Diagnostics

- `cgm_connection_status`
- `cgm_capabilities`
- `cgm_privacy_audit`
- `cgm_data_inventory`

### Reasoning

- `cgm_glucose_window`
- `cgm_daily_summary`
- `cgm_meal_response`
- `cgm_authorize_url`

## Prompts



## Resources

- `wellness-cgm-mcp://agent-manifest`
- `wellness-cgm-mcp://capabilities`
- `wellness-cgm-mcp://privacy-audit`

## Privacy

- OAuth tokens stored in env, never returned in tool output.
- Mock mode clearly tagged mock=true on every payload.
- CGM data is medical-record sensitive — agent rules enforce that.
- Defer insulin/medication decisions to a clinician.
- Profile and goals persist locally to ~/.delx-wellness/profile.json (mode 0600); shared with all 14 Delx Wellness connectors and never uploaded.

## Best for

- Per-meal glucose response coaching
- Dawn phenomenon detection
- Carb timing optimization for athletes
- Pre/post-workout glucose strategy

## Known limits

- v0.1 requires Dexcom only (Libre adapter in v0.2).
- Refresh-token rotation manual in v0.1.
- Not a replacement for clinical CGM workflows.

## Links

- GitHub: https://github.com/davidmosiah/wellness-cgm-mcp
- npm: https://www.npmjs.com/package/wellness-cgm-mcp
- Provider docs: https://developer.dexcom.com/docs/dexcomv2/overview