Audit Notifications with Cursor or Claude Code

Most teams can't answer a basic question about their own product: what notifications are you sending, and are they working?

This post walks through a full audit using Courier's CLI and MCP server from inside your coding environment. You'll need a free Courier API key. Every CLI command below also works as a typed MCP tool call in Cursor, Claude Code, or any agent that supports Courier's AI tooling.

npm install -g @trycourier/cli
export COURIER_API_KEY="your-api-key"

Inventory every template

courier notifications list --format json

This returns every template, its ID, and its state (published vs. draft). If you expected 15 and there are 40, the audit was overdue.

Pull the logs

Having a template doesn't mean it's firing. Check what's actually going out:

courier messages list --format json

# Count by delivery status
courier messages list --format json \
  --transform "results.#.status" | sort | uniq -c

UNDELIVERABLE and UNROUTABLE are your red flags. A handful is normal. A pattern means something is broken.

Trace a message end to end

Pick a message and walk its lifecycle:

# Status overview
courier messages retrieve --message-id "1-abc123" --format pretty

# Full event timeline: enqueued, sent, delivered, opened, clicked
courier messages history --message-id "1-abc123" --format pretty

# What was actually rendered and sent
courier messages output --message-id "1-abc123" --format pretty

Do this for a sample of each notification type. You're looking for where things fail (provider rejection? routing skip? preference block?) and whether the rendered content still makes sense. Stale copy, broken template variables, subject lines from 2024 -- this is where you find them.

For template-level inspection before data gets injected:

courier notifications content --notification-id "TEMPLATE_ID" --format json
courier notifications draft-content --notification-id "TEMPLATE_ID" --format json

If the published and draft versions have diverged, someone started editing and never shipped it.

Map the data

Every notification depends on data. Cross-reference what a template expects with what's actually being passed:

# Data payload from a specific send
courier messages retrieve --message-id "1-abc123" --format json

# What's on the user's profile
courier profiles retrieve --user-id "user-123" --format json

Build a simple map per notification type: what data it needs, where that data comes from, and whether it's consistently present. This is where you find silent failures -- the notification "works" but renders with blank fields because an upstream service isn't passing everything.

Break it down by tenant

For B2B products, notification health varies by customer org. list_messages accepts a tenant_id filter directly:

courier tenants list --format json
courier messages list --format json --tenant-id "acme-corp"
courier tenants retrieve --tenant-id "acme-corp" --format json

Look for tenants with outdated brand configs, concentrated delivery failures, or notification setups that haven't been touched since onboarding.

Check preference coverage

Notifications users can't opt out of are a churn risk.

courier preferences retrieve --user-id "user-123" --format pretty

Cross-reference with your template inventory. Every non-transactional notification should have a preference topic. If your marketing digest doesn't have a toggle, you're one annoyed user away from a full unsubscribe.

Build the hit list

Based on everything you've pulled, flag every notification that fits:

Kill it: UNDELIVERABLE/UNROUTABLE rate >10%. Zero engagement over 30 days. References features that no longer exist.

Revise it: Template variables rendering blank. Published/draft versions diverged. No preference topic assigned. Sent to channels without consent.

Batch or digest it: High volume, low urgency. Same notification type firing multiple times in minutes. Higher opt-out rates than other categories.

Investigate: Delivery rates that vary by tenant. No clear owner. Template untouched for 12+ months but still actively sending.

Take this list back to your team. The data backs up every recommendation.

Get started

• Get a free API key
• Install the CLI
• Set up the MCP server
• Install Courier Skills for best-practice guidance
• AI developer tools overview