MCP · Claude, ChatGPT, Cursor, your IDE
The Trackingplan MCP brings your tracking into Claude, ChatGPT, Cursor or your IDE: what fires, what broke, consent, and the spend from your ad accounts. Each person signs in as themselves, with no API key, and every answer shows the evidence behind it.
Ask it
The same agent answers through the MCP: it works from your real traffic, investigates before it answers, and shows the evidence. Two threads an analyst would open from Claude or Cursor.
What broke in my tracking this week?
Two things worth your time, nothing on fire.
1 · TikTok purchase fires twice on mobile web since v87 — inflating TikTok’s signal, GA4 unaffected.
2 · checkout_step lost coupon on web after v95 yesterday.
Everything else — 31 events, 14 providers, consent coverage — matches last week within normal variance.
Sample workspace: questions, answers and figures are illustrative. To try it on your own data, start your free trial.
Which campaigns lost match rate this week, and why?
Alert: Meta event match quality fell from 8.1 to 5.6 on Spring Sale – Prospecting and Retargeting – Cart since Tuesday. Those two campaigns are optimizing on a thinner signal; Google Ads is unaffected.
Cause: checkout now renders the confirmation page before the user object is in the dataLayer, so em goes out empty on 61% of purchases. fbp and fbc still arrive; this is a race, not a missing field.
Fix: push user before the purchase event, or fire the Meta tag on that object. EMQ should be back above 8 within a day of the change.
Sample workspace: questions, answers and figures are illustrative. To try it on your own data, start your free trial.
In the product
Connect · one address
In Settings & Members > Connect, copy your plan’s MCP endpoint and add it to Claude, ChatGPT, Cursor or your IDE as an HTTP MCP server. It asks you to sign in through your browser. No API key to share: everyone sees exactly what they see in the app.
Evidence · behind every number
A platform’s own MCP server shows your assistant the numbers that platform reports. Trackingplan shows it the data behind them: the hits your sites and apps send to GA4 and 170+ other providers, the dataLayer and the consent signals. Every answer shows the queries behind it.
Paid media · before you move budget
Ask where match rate is lost, where value is missing or where CAPI double-counts, with spend, clicks and reported conversions from eleven ad networks next to the traffic. Audits check what each pixel receives and end with the fix to ask for, so the signals you send get better. It never bids or changes a campaign.
Attribution · always on
Trackingplan keeps watching the click IDs, event_id and UTMs that connect a conversion to its ad, on real traffic. When one goes missing after a release, you ask why from the same chat where you were reading the campaign.
Changes · with your approval
From your assistant you can change the plan: create a draft event, delete an event with no hits in 30 days, change an event’s spec, or mute and unmute a warning. Nothing changes until you approve it.
Book a demo
This isn’t a sales call. It’s a chance to understand what matters most to you, discuss any data quality concerns you may have, and share practical ways Trackingplan could support your goals, if it makes sense.
FAQ
The MCP endpoint of your Trackingplan plan. Connect Claude, ChatGPT, Cursor or your IDE to it and you get an Ask Trackingplan tool. Behind that tool is the whole agent, with its data, its context and its standard of evidence, working from inside the assistant you already use.
Open Settings & Members > Connect, copy the plan’s MCP endpoint and add it to your tool as an HTTP MCP server. It asks you to sign in through your browser. You don’t need an API key: each person signs in with their own account. The steps are in the MCP endpoint docs.
What you see in the app: the hits your sites and apps send to GA4 and 170+ other providers, the dataLayer, the consent signals, the issues Trackingplan has found and your tracking plan. If an admin has connected your ad accounts, also spend, clicks and reported conversions per campaign from eleven ad networks (see Ad Platforms).
Your plan, only with your approval. You can ask it to create a draft event, delete an event with no hits in 30 days, change an event’s spec, or mute and unmute a warning; nothing changes until you approve it. It never bids or changes a campaign.
Because an AI agent can only reason about what it can see. A platform’s server hands your assistant the numbers that platform reports, broken events included. Trackingplan adds how each number was produced, so the assistant can tell a weaker campaign from a broken tag. Use both. We compare the platforms’ servers in analytics MCP servers.
The endpoint belongs to the plan, not to one person. If you turn it off, it closes for everyone who connects to it. All AI features in Trackingplan are optional, and plan admins can turn them on or off at any time.
More of Trackingplan