Direct Answer

AdRoll launched its Model Context Protocol (MCP) server in open beta in May 2026. Agencies can now pipe live campaign data directly into Claude, ChatGPT, and other AI tools via natural language. No exports. No cleanup. No delays.

Source: AdRoll Launches MCP Server to Make Campaign Insights and Actions Available Inside AI Tools

What Is MCP in Plain English

MCP (Model Context Protocol) is an open standard created by Anthropic. It's a bridge between AI models and your actual data sources. Before MCP, an AI had no access to your systems. You'd export a CSV, paste it into ChatGPT, get an answer, paste the next question. Slow. Error-prone. Now MCP lets Claude ask your systems directly. The system answers. Claude reads the answer. The manual disappears.

Think of it like this: instead of mailing a memo to the engine room asking what the reactor temperature is, you walk in and read the gauge yourself.

What AdRoll's MCP Server Does

AdRoll's implementation opens a secure data tunnel from your AdRoll account into Claude, ChatGPT, n8n, Microsoft Agent Copilot Studio, and Cursor. Your AI assistant can now:

  • Query campaign performance in natural language
  • Explore real-time account data
  • Draft campaigns (staged for review, not auto-published)
  • Analyze audience segments
  • Check attribution and conversion paths

This is read-and-write. Not read-only. Your AI can propose actions. You review before they go live.

The math is simple: what took 45 minutes manually now takes 45 seconds.

The Before/After

Before MCP:

  1. Log into AdRoll dashboard
  2. Export performance report (CSV)
  3. Download file
  4. Open ChatGPT or Claude
  5. Copy-paste the data
  6. Ask a question
  7. Get an answer based on yesterday's snapshot
  8. Repeat for each follow-up question

After MCP:

  1. Open Claude or ChatGPT
  2. Connect to AdRoll MCP
  3. Ask: "Which campaigns underperformed this week?"
  4. Claude queries live AdRoll data
  5. Get instant answer with current numbers
  6. Ask: "Which ones had the biggest drop in conversion rate?"
  7. Get pinpoint answer in seconds
  8. Ask: "Draft a new retargeting campaign for our e-commerce client targeting last-week visitors"
  9. Review the draft. Approve or refine. It stages in AdRoll ready to launch.

One workflow. Real data. No manual steps.

How Agencies Set It Up

The AdRoll MCP server requires:

  1. An active AdRoll or AdRoll ABM account
  2. An MCP-compatible AI tool (Claude, ChatGPT, Cursor, n8n, Microsoft Agent Copilot Studio)
  3. API credentials from AdRoll (API key, account ID)
  4. A few minutes to connect the MCP server to your AI platform

AdRoll's help center has the step-by-step setup. It's not complex. If you can paste an API key, you can do this.

The constraint isn't technical. It's permissions. Your API key determines what the AI can see and do. Set it right, and the AI has full read-write access to campaigns. Set it narrow, and the AI is read-only. Your choice.

The Watchstanding Principle

I spent years standing watch on nuclear submarines in the engine room. You don't ask the reactor for a written report every time you want to know its status. You walk to the gauges and read them. Directly. Immediately. That's the difference MCP makes.

Before MCP, every question to your ad system required a request, a wait, a report. Now you read the gauges yourself through Claude. Real-time visibility. Real-time action.

Owner-operators and agencies running at scale don't have time for email chains and manual exports. MCP saves that time. More important, it removes the lag between "I wonder if X is happening" and "I know X is happening."

Practical Use Cases

Performance Diagnostics:

A mid-market agency managing 15 client accounts. On Monday morning, the lead media strategist opens Claude and asks: "Summarize what changed in performance across all accounts this week." Claude pulls live data from AdRoll, compares it to the previous week, and flags the three accounts where conversion rates dropped. The strategist digs into those. Fifteen minutes. Decision made.

Without MCP, this takes an hour of manual pulling and pasting.

Campaign Creation at Scale:

An e-commerce agency needs to launch retargeting campaigns for 12 clients before EOD. Instead of logging into AdRoll 12 times, the media manager opens ChatGPT, specifies the client name, and asks: "Create a retargeting campaign draft for users who visited our product pages in the last 7 days." ChatGPT builds the draft, stages it in AdRoll. The manager reviews it. Approves. Next client. Done in 30 minutes instead of three hours.

Attribution Deep Dives:

A performance marketer notices a spike in conversions. She asks Claude: "Walk me through the attribution path for conversions in the last 48 hours. Which touchpoints had the highest influence?" Claude queries AdRoll's data, maps the funnel, shows which ads contributed. No manual query-building. No spreadsheet work. Answer in 90 seconds.

The Data Privacy Caveat

MCP is secure. Data flows between you and AdRoll, then to your AI tool. It's encrypted. It's not stored in ChatGPT or Claude's default memory (unless you have data retention on).

But you have a responsibility: your clients need to know you're running their data through AI systems. If you manage client accounts, get explicit consent. Document it. Make it part of your engagement agreement. The system is safe. The practice isn't transparent by default.

Sovereignty matters. Ownership matters. Make sure both are clear in writing.

What AdRoll's MCP Doesn't Do (Yet)

It's beta. Some things aren't included:

  • Cross-platform consolidation (only AdRoll data, not your Meta, Google, or TikTok accounts)
  • Automated actions without review (drafts only, not live deployment)
  • Historical data beyond what AdRoll retains (typically 2 years)
  • Custom metric creation inside the MCP interface (you're working with AdRoll's standard metrics)

These aren't blockers. They're just the edges of the tool. Understand them before you build a workflow around them.

What This Means for Agency Economics

Time savings compound into money saved.

If one media strategist saves five hours per week on manual reporting and campaign setup, that's 260 hours per year. At $75/hour loaded cost, that's $19,500 in recovered capacity per strategist per year.

Scale that across a five-person media team, and you've recovered nearly $100k in annual labor. Not profit. Capacity to do higher-use work. Strategy. Creative testing. Client retention.

The tool costs nothing to use if you already have AdRoll and Claude. The math is pure upside.

Action: Set Up Tomorrow

If you run an agency managing AdRoll accounts:

  1. Get your API key from AdRoll
  2. Open Claude or ChatGPT (your choice)
  3. Connect the MCP server (it takes five minutes)
  4. Test a simple query: "What are my top five campaigns by ROAS this week?"
  5. If it works, start building workflows

Start small. Use it for diagnostics first. Once you're comfortable, move to campaign drafting.

Don't wait for the perfect integration. Perfect is the enemy of done. Use it now.

Connection: Systems Beat Slogans

Every ad platform has a slogan: "Powerful Performance." "Intelligent Optimization." "Real-Time Insights."

MCP isn't a slogan. It's a system. Your AI talks to your ad system. Your ad system talks back. No middle layer. No delay. No translation.

That's how you actually get intelligent optimization. Not through promises. Through direct, real-time data flow.

This is the doctrine: build systems where the data flows fast, where the thinking happens in real time, where the human owns the decision. MCP is a tool that enforces that doctrine.

FAQ

Q: Do I have to use Claude? Can I use ChatGPT instead?

A: You can use ChatGPT, Claude, Cursor, Microsoft Agent Copilot Studio, n8n, or any tool that supports MCP. AdRoll's server works with all of them. Pick the one your team knows best.

Q: How much does the MCP server cost?

A: It's included free if you have an AdRoll account. No additional charge. You're paying for the Claude API calls (if using Claude) or ChatGPT Plus (if using ChatGPT), but the AdRoll integration itself is free.

Q: Can the AI make changes to campaigns without my review?

A: No. The MCP server drafts changes. Your team reviews them before they publish. Draft-first means you stay in control. The AI proposes. You decide.

Q: What happens if I revoke the API key?

A: The MCP connection stops working immediately. The AI can no longer access your AdRoll data. This is good. You can turn it on and off without breaking anything else.

Q: If I have multiple AdRoll accounts (one per client), can I run them all through one Claude session?

A: Yes, if your API key has access to multiple accounts. You can tell Claude "Use account X" in your prompt, and it will query the right account. Clean separation. No data leakage between accounts.

Q: Is there a learning curve for my team?

A: Minimal. Your team already knows how to write in English. That's all that's required. Ask Claude what you'd normally ask the dashboard. The MCP handles the translation to API calls and back.

Related Reading

*Disclosure: Jeff Barnes, MBA has no personal position in any company, fund, or platform named in this article. demg.ai has no current commercial relationship with any party mentioned. demg.ai provides marketing education and consulting services, not investment advice. Past performance does not guarantee future results.*