Weekly Marketing Reports with MCP: A Step-by-Step Playbook

A weekly marketing report pulls traffic, conversions and ad results into one page a manager can read in five minutes. In most small teams, a marketer spends part of every Monday exporting numbers from GA4 and ad platforms into a spreadsheet. With MCP (Model Context Protocol), an AI app such as Claude Code or Cursor can query those platforms directly through MCP servers, so you ask for the numbers in plain English and review a draft instead of copying cells.

This playbook uses only official servers that Google, Meta and Microsoft document. It is part of our MCP playbooks series. Everything below was checked against the vendors' docs on 28 October 2026. eSEOspace did not install or test these servers.

What you need

You need one MCP client and two to four servers. Claude Code, Claude Desktop and Cursor are all named as clients in Google's Analytics MCP docs.

GA4 options, checked. Google's Analytics MCP page lists a local server published on PyPI as analytics-mcp. It needs the Admin API and Data API turned on in a Google Cloud project, the Viewer role on the property and the analytics.readonly scope. The page lists no hosted remote URL. Its tools include get_account_summaries, get_property_details, list_google_ads_links, run_report, run_realtime_report, run_funnel_report, get_custom_dimensions_and_metrics and run_conversions_report.

The GitHub README is titled "Experimental." It does not list run_conversions_report, and it adds a create_property_annotation tool that needs the --allow-write flag and the analytics.edit scope. Confirm the tool list in your client after you connect.

Search Console. Google does not publish a Search Console MCP server as of October 2026. If you already pay for Ahrefs, its tool categories page lists a gsc group covering "keywords, pages, positions, and performance history." Ahrefs requires a paid plan from Lite up.

The workflow

1. Set up read-only access first. Connect GA4 with the Viewer role and analytics.readonly. Connect Google Ads with a login that sees only the accounts you report on. For Meta, grant ads_read plus the required ads_mcp_management, and leave ads_management off.
2. Confirm the property and accounts. Ask the AI to run get_account_summaries in GA4 and list_accessible_customers in Google Ads. Write down the property ID and customer IDs the report should use, so the AI does not guess.
3. Pull last week's GA4 numbers. The AI calls run_report for sessions, users and key events by channel, comparing the last 7 days with the 7 days before. Use run_funnel_report if you track a checkout or lead funnel.
4. Pull paid search results. The AI uses the Google Ads search tool, which runs Google Ads Query Language (GAQL) queries. If a query fails, get_resource_metadata tells the model which fields a resource such as "campaign" supports.
5. Pull paid social results. Meta's overview lists "comprehensive reporting" as a tool category. Ask for spend, results and cost per result by campaign for the same 7 days.
6. Add behavior signals (optional). Clarity's query-analytics-dashboard tool can answer questions about rage clicks, dead clicks or script errors. Microsoft's Data Export API docs limit this to 10 requests per project per day and the last 1 to 3 days of data.
7. Draft the report. Ask the AI to build one table per channel and a short "what changed" list. Tell it to show the numbers it pulled next to each claim.
8. Review and send. A person checks totals against the GA4 and ads interfaces, edits the commentary and sends it. Save the prompt so next week's run is the same.

Example prompts

• "Use get_account_summaries to list my GA4 properties. Then run a report for property [ID] showing sessions and key events by session default channel group for the last 7 days versus the previous 7 days."
• "Run run_funnel_report for the steps view_item, add_to_cart and purchase for last week. Show drop-off at each step."
• "Using the Google Ads search tool, list cost, clicks, conversions and cost per conversion by campaign for customer [ID] for the last 7 days. If a field is invalid, check get_resource_metadata first."
• "From the Meta ads server, report spend and results by campaign for ad account [ID] for the last 7 days. Do not create or change anything."
• "Ask Clarity which pages had the most rage clicks in the last 3 days, then put all of this into a one-page summary with a table per channel."

Guardrails

• Read-only first. GA4 and Google Ads are read-only by default. Keep --allow-write off for GA4 unless you want the AI to write annotations.
• Meta needs extra care. Meta's server can "create and edit ads, ad sets, and campaigns." For reporting, use a token with ads_read only, and keep approval prompts on in your client.
• Protect credential files. The GA4 and Ads servers sign in with Application Default Credentials. That JSON file opens the same data as the Google account behind it.
• Data goes to the model. Google's Ads README warns that the server "will expose your data to the Agent or LLM that you connect to it." Check that this fits your privacy policy.
• Keep humans on judgment. Explaining why numbers moved, and what to do about it, stays with a marketer. The AI can be wrong about cause and effect.

Read the MCP security hub before you share one setup across a team.

What to measure

Track whether the playbook saves time without hurting accuracy:

• Time from "start report" to "report sent," before and after.
• How many numbers a reviewer had to correct each week, and why.
• How often a tool call failed or returned the wrong account or date range.
• Clarity API requests used per day, so you stay under the cap.
• Whether readers act on the report, such as fewer follow-up questions from managers.

Keep these in a simple log for four to six weeks before you decide whether to expand.

Next step

If you want this report running for your team, with the right scopes and a reviewed prompt, our MCP marketing service can set it up. For more marketing workflows, read MCP for marketing teams.

Frequently asked questions