Marketing Attribution MCP: Which Channel Actually Pays You

Every analytics tool can tell you which channel sent the most visitors. Almost none can tell you which channel sent the most revenue — and fewer still can tell you which one still looks good after churn.

A marketing attribution MCP puts that answer inside your AI assistant. Instead of exporting traffic data from one tool, revenue from another, and joining them in a spreadsheet, you ask a question and the assistant reads both sides directly.

This post covers what a marketing attribution MCP exposes, why joining traffic to revenue is the hard part, the questions it makes answerable, and where the numbers stop being trustworthy. It is written for SaaS founders running acquisition across more than one channel.

What is a marketing attribution MCP?

A marketing attribution MCP is a server that exposes traffic, signup, and revenue data as connected read-only tools an AI assistant can query — so it can trace a channel from first visit through signup to actual subscription revenue. MCP is the Model Context Protocol, an open standard for connecting AI clients to external systems.

The word doing the work is connected. Traffic data alone is a vanity metric. Revenue data alone cannot tell you where customers came from. Attribution is the join between them, and it is precisely the join that is normally missing.

The join that nobody has

Answering "how much MRR did Product Hunt actually generate?" requires three systems that were never designed to talk to each other:

1. Site analytics — who arrived, and from where.
2. Your signup flow — which of those visitors became accounts.
3. Your billing provider — which of those accounts became revenue, and which churned. (The full visits → signups → MRR funnel walks through each stage.)

Most teams do this join by hand, quarterly, badly. It is slow enough that nobody does it monthly, which means channel decisions get made on signup counts instead of revenue. We unpack the join itself in marketing attribution for SaaS revenue.

In practice, the tooling gap is why so many teams over-invest in channels that convert well and retain poorly. A channel that drives cheap signups who churn in month two can look like your best performer right up until you measure revenue instead of registrations.

What the tools expose

Chartsy's MCP server carries the attribution half of the product — Growth — into your AI client. The relevant tools:

• Traffic sources — where traffic came from, with signups and signup rate per group, over any date range. Groupable by source, medium, campaign, country, referrer, or landing path.
• Site traffic over time — visits, unique visitors, and signups per day, week, or month.
• Your tracked sites — the list of sites on your account, so a question can be scoped to one.
• The full business snapshot — the whole business over one shared window, including the site-traffic half alongside MRR, churn, and customer data.

That last one is the important one for attribution work, because it puts traffic and revenue in the same response over the same time window. The assistant does not have to reconcile two differently-scoped queries.

First touch and last touch, both

Attribution collapses if you only keep one. Growth records both:

• Last touch — the source credited with the visit immediately before the action. Answers "what brought them back this time?"
• First touch — the very first source ever recorded for that visitor. Answers "what made them aware of us at all?"

Reading them together is the point. Last touch over-credits channels sitting close to the decision — direct visits and branded search — while first touch ignores whatever actually closed the deal. An assistant with access to both can tell you which channels start relationships and which finish them.

Questions it can finally answer

1. "Which traffic source has the best signup rate this quarter, and how does it compare to last?"
2. "Which landing paths bring in visitors who actually sign up?"
3. "Compare first-touch and last-touch credit for my top five sources."
4. "Did the campaign we ran last month produce signups, or just traffic?"
5. "Which countries send traffic that converts, and which just send traffic?"
6. "Draft a channel performance summary for the team from the last 90 days."

The shift is from reporting to arbitration. You are no longer asking what happened; you are asking which of two channels deserves next month's budget.

Where the numbers stop being trustworthy

Any attribution system has limits, and an honest one tells you where they are.

• Direct is not a channel. A visit with no UTM parameter, click ID, referral parameter, or usable referrer is labelled direct and counted as such. Growth does not force it into a bucket to make totals look tidier — which means direct is genuinely unclassifiable traffic, not a category to optimise.
• Untagged links vanish into direct. A newsletter shared without utm_source arrives with nothing to classify — see connecting UTM tracking to revenue.
• Cookie-less visitors are attributed visit by visit. First-touch and last-touch attribution are held in cookies. For a visitor Chartsy stores nothing for — an EEA or UK visitor on a site with no consent banner wired up, for instance — there is no first touch carried forward.
• Revenue attribution needs billing connected. Traffic and signups work as soon as tracking is installed; MRR and churn by channel require a connected Stripe or Paddle account.

Undercounting on purpose is the right trade. A system that guesses produces numbers that feel complete and are quietly wrong. One that says "direct" when it does not know produces numbers you can act on.

Conclusion

A marketing attribution MCP is worth setting up when your channel decisions have outgrown signup counts. By exposing traffic, signups, and subscription revenue as connected tools, it lets your AI assistant answer the question every founder actually has — which channel is worth continuing — without a manual join across three systems.

The caveat is the same as with any attribution: the answers are only as good as your tagging, and unclassifiable traffic should stay unclassified.

See how Growth attribution works → · Set up the Chartsy MCP server →

Related reading

• Google Analytics MCP vs revenue analytics MCP
• SaaS analytics MCP: ask Claude about your MRR and churn
• What is an analytics MCP?

Frequently asked questions

What is a marketing attribution MCP?

It is a Model Context Protocol server that exposes traffic, signup, and revenue data as connected read-only tools, so an AI assistant can trace a marketing channel from first visit through signup to real subscription revenue without a manual export and join.

Can an AI assistant calculate revenue by channel on its own?

Not reliably from separate exports. It needs traffic and revenue joined on a persistent visitor identity, over a consistent time window. That join happens in the analytics platform; the MCP server just makes the result queryable.

Does attribution work across subdomains?

Yes, when cookies are available. First-touch and last-touch cookies are shared across subdomains, so someone who lands on your marketing site and signs up on an app subdomain keeps their original channel instead of appearing as an internal referral.

Why is so much of my traffic showing as direct?

Usually untagged links, a redirect stripping the query string before the tracked page loads, or the tracking script missing from some entry pages. Direct means no UTM parameter, click ID, referral parameter, or usable referrer was present.

Do I need billing connected for attribution to work?

Not for traffic. Visits and signups record as soon as tracking is installed. Revenue and churn by channel require a connected Stripe or Paddle account, since that is the data signups are matched against.