Matomo and AI: Is there an MCP server for Matomo?

Matomo is web analytics software that runs as a cloud service or self-hosted (On-Premise). If you want an AI assistant such as Claude or ChatGPT to work with these reports, the first question is: is there an MCP server for it?

In short

Yes. Matomo offers an official MCP server. According to Matomo, it is included in Matomo Cloud. For On-Premise, it is a free plugin. Access is off by default and runs with the rights of the signed-in user.

What Matomo offers itself

The MCP server. In its guide, Matomo describes it as a mediation layer between an AI tool and the analytics data. It accepts structured requests, forwards them to the Matomo Analytics API and returns the data. According to Matomo, the connected AI tool alone does the interpretation and writes the answer. The endpoint has the form index.php?module=API&method=McpServer.mcp&format=mcp. According to the plugin description, the tools cover sites, reports, report metadata, goals, segments and dimensions. An administrator can also allow access to the raw Matomo API.

Switching it on. A superuser enables the server under Administration > System > General settings > MCP Server. According to Matomo, this holds for Cloud and On-Premise. The same place sets the highest privilege level allowed through MCP, and the raw API access. Matomo names “no limit” as the default for the privilege level and “no API access” for the raw access.

Sign-in. According to Matomo, every request needs a valid Matomo API token or OAuth 2.0. The token inherits the rights of its user. From Matomo 6, you can choose an access level when you create it, which cannot exceed the user’s rights. With OAuth 2.0, the scopes of the OAuth client apply. Matomo names limits: not all features of the OAuth specification are implemented. According to Matomo, clients that require fully specification-compliant endpoints may not work. The plugin description, in contrast, names automatic discovery of the OAuth server when the OAuth 2.0 plugin is active and the client supports MCP authorization. According to Matomo, OAuth access tokens last 3600 seconds by default, and refreshing them currently has to be done by hand. Matomo recommends a dedicated user with few rights for MCP.

Plans and plugin. According to Matomo, the MCP server is included in Matomo Cloud. For On-Premise, the Marketplace lists the plugin “MCP Server” as a free download under GPL v3+. The source code is at matomo-org/plugin-McpServer. For Matomo 5.x, the Marketplace page names version 5.8.0-rc1 as the minimum (6.0.0-b1 for 6.x), and PHP 8.1. According to the Marketplace, the plugin is not available for Matomo for WordPress. We did not evaluate the prices of Matomo Cloud.

Claude, ChatGPT and other assistants. Matomo publishes guides for several clients:

• Claude Code: with a token in the Authorization header (according to the guide, also with an OAuth access token).
• Claude Desktop: as a custom connector with OAuth 2.0. According to Matomo (updated August 2026), Claude does not support a token here.
• ChatGPT on the web: as a custom MCP server with OAuth. According to Matomo, you need a ChatGPT account with developer mode enabled.
• OpenAI desktop app and Codex CLI: with a token or OAuth 2.0.
• Mistral AI: with a token.
• Microsoft Copilot Studio: through a custom connector in Power Apps, with a token.
• Other clients: according to the general guide, possible if they connect to a remote MCP server with a token or OAuth.

Whether your Claude or ChatGPT plan allows custom connectors is not stated by Matomo. Anthropic and OpenAI decide that.

An API. The Analytics API delivers all reports. According to Matomo, it also manages sites, users, permissions and goals, including deleting. It uses an API token or OAuth 2.0.

AI in the product. According to Matomo, “Ask Matomo” arrives in November 2026. It is an optional AI chat for signed-in users. A superuser must switch it on first. On Matomo Cloud, it runs through Amazon Bedrock from AWS, according to Matomo. Matomo says it uses only EU regions where the model allows it. According to Matomo, models with global or cross-region processing may process requests outside the EU. The AI goal recommendations can be used in Matomo Cloud, according to Matomo, and are announced for On-Premise. There is also a referrer channel, “AI Assistant”, which reports visits from AI tools such as ChatGPT, Claude and Gemini separately. That is web analytics about AI traffic, not an AI assistant.

Data location. According to the pricing page, the cloud version stores data in Frankfurt, Germany. With On-Premise, you choose the location yourself. According to the privacy policy, Matomo Cloud data and backups are stored in Europe. According to that policy, Matomo Cloud is operated by InnoCraft Limited, a company registered in New Zealand. We found no statement on where the Matomo Cloud MCP endpoint runs in detail. According to Matomo’s privacy notice, the AI provider (for example OpenAI or Anthropic) is a recipient that processes the data independently of Matomo. Matomo recommends listing it in your privacy policy and doing a privacy assessment beforehand.

What the community and third parties offer

There are several free servers. FGRibreau/mcp-matomo (MIT, written in Rust) generates tools from the API of your Matomo instance. According to its README, it is read-only by default. The switch --enable-writes adds tools for goals and funnels, including deleting. kitconcept/matomo-mcp (MIT, Python) uses an API token to reach the Reporting API, according to its README. The repository’s last commit was in November 2025.

We name commercial offers separately. According to Openmost, the Matomo agency runs a hosted MCP server with about 73 tools and OAuth sign-in through an Openmost account. There is also a client, a small Node.js program for Claude Desktop. According to Openmost, it is free, but it needs an Openmost token, which according to the README comes on request through a contact form. According to the README, the Matomo address and Matomo token go to the Openmost server with each request. According to Openmost, the repository has no licence file. We found no prices for the hosted server. Windsor.ai offers a connector that brings Matomo data into Claude, according to its own page. We did not check what these pages leave unproven about rights and data paths.

What this means for your company

• The MCP server stays off until a superuser enables it. Review the privilege level and raw API access first.
• Matomo advises a dedicated user that has only the required rights. Matomo warns about prompt injection through values such as page titles and advises preferring read access.
• What reaches the AI assistant falls under its terms. Add the provider to your privacy policy and check whether a data processing agreement is needed.
• Claude Desktop needs OAuth 2.0. With On-Premise, that includes the OAuth 2.0 plugin.