Open source · MIT · by Analitik İşler

Newv0.8.0 is on npm as @analitikisler/ai-tag-debugger →

AI Tag Debugger tests your analytics and ad tracking. Open source.

An AI model clicks through your site like a visitor. Rules check every GA4, Google Ads, Meta and TikTok hit against your measurement plan and Consent Mode. Run it in your terminal, in CI after each deploy, or from your AI assistant.

Install

$ npm install -g @analitikisler/ai-tag-debugger

$ npx -y @analitikisler/ai-tag-debugger run --site https://staging.yourshop.com --journey add_to_cart

"ai-tag-debugger": { "command": "npx", "args": ["-y", "@analitikisler/ai-tag-debugger", "mcp"] }

Node 20 or newer. Free and MIT licensed.Then follow the quickstart →

Checks the tags you already run

GA4Google AdsMeta PixelTikTok PixelX Pixel

LinkedInMicrosoft AdsPinterestSnapchatClarityHotjarDoubleClickAdSense

Pick any of them with --providers.

One real run

Three planted bugs. One command found all of them.

We built a demo shop with known tracking bugs and ran two journeys on desktop and mobile. This run used rules only, with no AI model, so anyone can repeat it for free.

~/demo-shop · ai-tag-debugger

$ ai-tag-debugger run --config examples/demo.config.json --scripted --no-ai ▶ checkout (mobile)  Accept cookies  Open the first product  Add to cart  Open the cart  Go to checkout  …  Place the order  ✓ All scripted steps completed. · 20 tracking hits captured ✗ broken   GA4 purchase sent without currency, value (mobile)✗ broken   Meta Purchase sent without value, currency (mobile)! risk     Meta PageView fired before the visitor accepted cookies! warning  add_to_cart sent 2 times by one action 2 broken · 1 risk · 1 warning · 4 okReport: reports/demo/report.html

Findings by severity

broken

risk

warning

purchase sent to GA4 and Meta without value and currency on mobile
PageView sent to Meta before the visitor accepted cookies
add_to_cart sent twice by one click

2 journeys · desktop and mobile · 0 model callsRun it yourself →

How it works

Four steps, one command. AI browses and explains. Rules do the checking.

Browse

Instead of clicking through checkout by hand

An AI model follows a goal in plain words, or replays fixed steps with no model at all.

--journey purchase

Capture

Instead of the Network tab

Every hit and dataLayer push, tagged with the step that caused it. Test hits are answered locally, so your real accounts stay clean.

--providers ga4,meta,tiktok

Check

Instead of a spreadsheet of expected events

Rules compare hits with your measurement plan: missing events and parameters, duplicates, consent and personal data.

--events purchase,add_to_cart

Explain

Instead of writing the ticket yourself

The model writes the likely cause and the fix for each problem, in English or Turkish.

--lang tr

Quickstart

From install to a report in five commands.

~/my-shop · install

$ npm install -g @analitikisler/ai-tag-debugger
$ npx -y playwright@1.56.1 install chromium

The tool runs on Node 20 or newer. Playwright drives a real Chromium browser.

The report

Every hit, every step. The worst problem first.

One HTML file with four tabs: Overview, Findings, Journeys, and Scope and plan. Send it to a developer or a client. It needs no server and runs no scripts.

reports/demo/report.html#journeys

The Journeys tab of an AI Tag Debugger report: a step list on the left and the selected step's events on the right, with the Meta PageView flagged for firing before consent

Open the full demo report →

tabs per step: Events, dataLayer, Network

platforms you can choose

ready-made journeys

test hits sent to your real ad accounts

report.htmlSummary, findings with cause and fix, full timeline
timeline.csvOne row per hit, for Excel or Google Sheets
report.mdFor a pull request comment
capture.jsonEverything captured, for your own analysis
[redacted]Emails, phones and user ids are masked before they're stored

AI assistants

Ask your AI assistant to check the tracking. No API key needed.

AI Tag Debugger is also an MCP server, so it works with Claude, ChatGPT, Gemini and other assistants that support MCP. The assistant's own model drives the browser and explains the findings in the chat.

{
  "mcpServers": {
    "ai-tag-debugger": {
      "command": "npx",
      "args": ["-y", "@analitikisler/ai-tag-debugger", "mcp"]
    }
  }
}

{
  "mcpServers": {
    "ai-tag-debugger": {
      "command": "npx",
      "args": ["-y", "@analitikisler/ai-tag-debugger", "mcp"]
    }
  }
}

# Add it as an MCP server that runs this command
npx -y @analitikisler/ai-tag-debugger mcp

# Install the browser once
npx -y playwright@1.56.1 install chromium

# 1. Expose a local port
cloudflared tunnel --url http://localhost:8787

# 2. Start the server in HTTP mode
npx -y @analitikisler/ai-tag-debugger mcp --http --port 8787 \
  --public-url https://<your-tunnel-address>

# 3. Add the printed /mcp/<secret> URL as a connector

export ANTHROPIC_API_KEY=...   # or OPENAI_API_KEY, GEMINI_API_KEY

ai-tag-debugger run --site https://staging.yourshop.com \
  --journey purchase --viewport mobile

# Mistral, DeepSeek, Groq, OpenRouter, a local Ollama...
ai-tag-debugger run --provider openai-compatible \
  --base-url http://localhost:11434/v1 --model qwen3 ...

Analitik İşler Cloud

AI Tag Debugger, run for you. On a schedule and after every release.

AI Tag Debugger stays free and open source. Analitik İşler Cloud is our hosted service on top of it, with history, alerts and many sites in one place.

	AI Tag DebuggerOpen source · Free	Analitik İşler CloudHosted · Contact us
Testing
AI and rules-only runs	✓	✓
Measurement plan and consent checks	✓	✓
Report in English or Turkish	✓	✓
Running it
After each deploy	Your CI (GitHub Actions example)	✓ Deploy webhooks
On a schedule	—	✓
From ChatGPT or Gemini	Your own tunnel	✓ Hosted MCP URL
Teams and agencies
Run history and alerts	—	✓
Many client sites in one place	—	✓
Consent Check and Measurement Plan drafts	—	In development

Get access to Cloudor write to hello@analitikisler.com

Try it on your staging site before the next release.

$ npx -y @analitikisler/ai-tag-debugger run --site https://staging.yourshop.com --journey purchase

Quickstart Get access to Cloud