Google Analytics MCP: No-Code SetupPorter Metrics connects Google Analytics 4 to Claude and ChatGPT through one hosted MCP server. You ask about sessions, landing pages and key events in a chat, then fix tracking settings after you approve. Porter reads Search Console and your ad platforms through the same server, so traffic and its cause share one answer.
One MCP server: Porter connects Google Analytics 4 to the AI clients, fields, metrics, dimensions and workflows marketers already use. You paste one server URL and sign in with Google once. Every client you add then reads the same 288 fields, and the same server reaches the sources in all MCP connectors.
Only one AI app?
Objects: properties · data streams · key events · custom dimensions · custom metrics · audiences · users
Traffic: Sessions · Total users · New users
Engagement: Engaged sessions · Engagement rate · Average session duration
Conversion: Key events · Event count · Purchase revenue · Ecommerce purchases
Definitions for each one are in the GA4 field definitions.
Acquisition: Session default channel group · Session source / medium
Content: Landing page + query string · Page path · Page title · Event name
Audience: Device category · Country
Time: Date · Week · Month and year
Rule of thumb: GA4 updates through the day, so prompts ask for periods ending two days ago.
The same Google Analytics connection also sends data to Data Studio (formerly Looker Studio), BigQuery and Google Sheets, at no extra cost.
Start from the free Google Analytics 4 dashboard templates.
Porter’s Google Analytics connection reads first and asks before any change. A Viewer role is enough to report, an Editor role lets Claude change settings, and every change waits for your approval. Actions GA4 can’t undo, like removing a key event, are flagged as permanent, so you confirm them before they run.
Read-only first: ask about sessions, pages and key events without changing anything in the property. Reporting runs apart from settings changes.
Unlimited reporting queries: ask as many questions as you need. Porter works within Google’s Data API quota of 200,000 core tokens per property per day, so you never track tokens yourself.
Approved Google sign-in: connect with your Google account. You never create a Cloud project, a service account or a credentials file.
Your role decides the scope: Google Analytics roles (Viewer and Editor) carry over. Viewer reads reports; Editor is needed for key events, custom definitions and streams.
Changes after you approve: Claude shows the key event, dimension or audience it will create. Nothing reaches GA4 until you say yes.
Permanent actions flagged: removing a key event, unlinking Google Ads, deleting a data stream or removing a user can’t be undone in GA4. Porter flags each one and asks you to confirm.
Official API: Porter uses Google’s Analytics Data and Admin APIs, never a browser session clicking through GA4.
Revoke access: disconnect Google Analytics from your Porter account whenever you want. Access stays tied to that sign-in, not a key on your laptop.
Setting up Porter’s Google Analytics connection takes one server URL and a Google sign-in. There is no Python, no Cloud project and no server to run. Claude and ChatGPT each have their own place to paste the URL, then a first question about last month’s sessions confirms the property is connected.
Use the Claude plugin for the guided route, since Claude keeps the Porter plugin synced when new tools ship.
portermetrics/porter.portermetrics/porter.First question to paste:
For my GA4 property, show Sessions, Engaged sessions, Engagement rate and Key events by Week for the last 4 full weeks, ending two days ago.Claude also documents the manual route in Anthropic’s guide to custom connectors using remote MCP. For screenshots and Claude Code, follow how to connect Google Analytics 4 to Claude, or see the Google Analytics 4 connector for Claude for Claude-only use cases.
Click-by-click Claude plugin demo:
Porter’s walkthrough connects GA4 to Claude without code. You sign in with Google in Porter, add Porter’s MCP URL to Claude as a custom connector, then ask a first traffic question. The steps match the list above.
Use manual setup when Claude asks for a custom connector instead of a plugin.
Porter.https://mcp.portermetrics.com/mcp.Manual setup demo:
Use the Porter Metrics app for the fastest ChatGPT route.
OpenAI explains app permissions in OpenAI’s guide to connected apps in ChatGPT. Screenshots are in how to connect Google Analytics 4 to ChatGPT, and ChatGPT-only use cases sit on the Google Analytics 4 connector for ChatGPT.
Click-by-click ChatGPT app demo:
Use Connect MCP for the full Porter setup in ChatGPT: connector plus skill. Plan requirements are in OpenAI’s guide to connecting custom MCP servers in ChatGPT.
Porter Metrics.https://mcp.portermetrics.com/mcp as the MCP Server URL.https://mcp.portermetrics.com/skill.zip.skill.zip to your ChatGPT project.
Pick your coding agent. Each one uses the same server URL and a Google sign-in, so no API key or credentials file sits in a config.
--scope user to use Porter in every project.claude mcp add --transport http porter https://mcp.portermetrics.com/mcp/mcp./mcp shows porter as connected. Ask: “Which Google Analytics accounts can you see?”mcp.json.{
"mcpServers": {
"porter": { "url": "https://mcp.portermetrics.com/mcp" }
}
}Ctrl/Cmd+Shift+P) and run MCP: Add Server → HTTP.https://mcp.portermetrics.com/mcp, name it porter and choose Global. Or add it to .vscode/mcp.json:{
"servers": {
"porter": { "type": "http", "url": "https://mcp.portermetrics.com/mcp" }
}
}codex mcp add porter --url https://mcp.portermetrics.com/mcpcodex mcp login portercodex mcp list shows porter as logged in. Start codex. Ask: “Which Google Analytics accounts can you see?”The same URL works in Windsurf, Cline, Zed and any client that supports remote MCP servers over HTTP.
Google’s server has no hosted URL. You install it on your own machine from Google’s analytics-mcp repository on GitHub and run it with pipx.
Use Google’s server if you’re a developer who wants free, local, read-only access and is happy to manage credentials. Use Porter’s URL (https://mcp.portermetrics.com/mcp) when you want GA4 next to Search Console and ad spend, across client properties, with dashboards and approved fixes. The two can run side by side.
Porter’s URL goes into the server URL field of any MCP client; Google’s runs as a local command.
Porter’s Google Analytics skills are working prompts, not tutorials. Each one turns live GA4 data into a report, a diagnosis or an approved tracking fix, and you can paste it as written. Reporting and research prompts only read the property. Automation prompts propose a change and wait for your approval.
Marketers who need the why behind the traffic. Porter’s Google Analytics connection fits four groups. Agencies report on many client properties, and in-house marketers own the website numbers. SEO and content marketers track organic and AI traffic, and analysts keep tracking clean. Each group below gets four jobs it can hand over.
You’re a developer, read-only access is enough, and you’re happy to run Python with a Cloud project. Google’s server is free, open source and enough for that job.
Google publishes an open-source Google Analytics MCP server that runs on your own machine. Google’s Analytics MCP setup guide describes it as read-only and lists its requirements (as of November 2026). It suits developers, and it can run next to Porter in the same AI client.
Google’s server gives your AI read access to GA4 from your laptop. Porter gives it your marketing operation: GA4, Search Console, ad spend, dashboards and approved fixes from one hosted URL.
The basics are the same on both: Google’s official Analytics APIs, your existing GA4 access, every report field GA4 exposes, and a server any MCP client can use. Here is where they part.
| Porter | Google’s official MCP (as of November 2026) | |
|---|---|---|
| Same job, different reach | ||
| Setup | Google sign-in, one URL | Python 3.10+, a Cloud project with the Admin and Data APIs, Application Default Credentials |
| Where it runs | Hosted by Porter | On your machine, through pipx |
| Changes to GA4 | Key events, custom dimensions and metrics, audiences, streams, Google Ads links, user access | Property annotations only, behind an --allow-write flag |
| Approval controls | Every change shown first and approved | Read-only by default |
| Clients in its docs | Claude, ChatGPT, Gemini, n8n, any remote MCP client | Antigravity, Claude Desktop, Claude Code |
| Price | Included in every Porter plan, the free plan too | Open source |
| Only one side | ||
| Search Console and other channels | 30+ sources: Search Console, Google Ads, Meta Ads, HubSpot and more | GA4 only |
| Cross-channel in one conversation | Yes | Not in its docs |
| Blending sources into one number | Yes | Not in its docs |
| Live dashboards at their own link | Yes | Not in its docs |
| Scheduled reports and alerts | Yes | Not in its docs |
| Attribution to CRM revenue | Yes | Not in its docs |
| Data Studio, Sheets, BigQuery | Yes | MCP clients only |
| Skills: whole tasks already written | Yes | You write the prompt |
Use Google’s own if you’re a developer, need GA4 only, and read-only access is enough. Everything under Only one side is what happens when the job grows past that.
Plenty of servers rank for this search, and they split three ways. The job tells you which one you’re looking at.
Porter reads, reports and fixes tracking from one hosted server, on every plan.
The Google Analytics MCP is included in every Porter plan, the free-forever plan too. Reporting is billed per connected account, and each GA4 property counts as one account.
Reporting price for connected accounts, in USD. AI chats, queries and dashboards are unlimited. Annual billing is 10 months for 12.
The more accounts you connect, the lower the price per account. See the full calculator on Porter pricing.
Yes. Google publishes an open-source server you run locally, and Porter hosts one at https://mcp.portermetrics.com/mcp. Both read GA4 through Google's official APIs. Porter adds a Google sign-in instead of a Cloud project, Search Console and ad data in the same chat, and key event fixes after approval.
Porter's server URL is https://mcp.portermetrics.com/mcp, and you paste it into Claude, ChatGPT or any MCP client. Google's own server has no hosted URL. You install it from Google's analytics-mcp repository on GitHub and run it on your machine with Python and your own credentials.
Yes. Google's Google Analytics MCP server is open source and runs locally with Python 3.10 or higher, a Cloud project and Application Default Credentials. Google describes it as read-only. Its only write tool, property annotations, needs an --allow-write flag (as of November 2026).
Yes. Google Analytics 4, or GA4, is the current version of Google Analytics, so both searches point to the same kind of server. Porter's Google Analytics connection reads GA4 properties, and its fields use the names you see in GA4, like Sessions and Key events.
MCP is the Model Context Protocol, an open standard that lets AI apps like Claude read data from other tools. MCP is not the agent. Claude and ChatGPT are the agents, and MCP is how they reach your GA4 property through a server like Porter's.
Install the Porter plugin from Claude Settings → Customize → Plugins. Add the portermetrics/porter marketplace, install Porter Metrics, connect the Porter connector and sign in with Google. Then ask about last month's sessions. Use the manual custom connector only when Claude asks for one.
Porter's MCP costs nothing on any Porter plan, and it reaches GA4 through Google's official APIs within Google's property quotas. What you pay Porter for is Reporting, per connected property, and Automation credits when Claude changes settings. The free plan covers 30 days of history and 100 credits a month.
Connect Google Analytics to Claude or ChatGPT and ask why traffic moved, which pages convert and what to fix today.
Start free: unlimited GA4 properties, no credit card.
For agencies: every client property answers in one chat, and the Monday report is a link.
For SEO and content marketers: Search Console and GA4 finally sit in the same answer.
For analysts: broken key events reach you as a fix to approve, not a surprise in the report.