Hosted vs Self-Hosted Matomo MCP: Which Setup Fits?

A Matomo MCP server connects an AI client to tools that query Matomo. You can operate a server associated with your own installation or use a hosted service. The right choice depends on which reports and actions you need, who operates the endpoint and which data path your organisation approves.

The official Matomo MCP route

Matomo publishes an official MCP guide and server plugin. On an On-Premise installation, follow its installation and configuration requirements. Check the guide for your hosting environment and version rather than assuming every deployment works the same way.

This route is worth evaluating when you want to operate the Matomo-side integration yourself and can maintain its access controls and connectivity. You still need to review what data your chosen AI provider receives; running a connector yourself does not keep the model’s processing inside your infrastructure automatically.

A hosted Matomo MCP service

A hosted service provides the remote MCP endpoint and queries your existing Matomo. With Savigny, you connect a public HTTPS Matomo installation, choose the websites to share and authorize a compatible AI client through OAuth. No Matomo plugin is required.

The endpoint is:

https://mcp.savigny.io/mcp

Savigny is read-only. Its current seven tools cover websites, visits, pages, traffic sources, referring websites, countries and a month comparison. It does not expose arbitrary Matomo API methods, configuration writes, goals, events or conversion reports. It currently supports one Matomo connection per user.

Other hosted connectors have different scopes. Compare Openmost and Savigny if you need to evaluate broader reporting and configuration capabilities alongside a focused traffic-reporting service.

Compare the actual data path

Write down which systems receive credentials, report results and conversation content. For Savigny, the Matomo token is stored encrypted on the server, the service retrieves authorised reports, and your AI client receives the requested report results. A hosted bridge is a further service in the path; read its privacy information and your AI provider’s policies.

A Matomo installation available only through a VPN or private network will not work with Savigny. If that is a requirement, evaluate an architecture with the necessary approved connectivity rather than opening access without a plan.

Compare actions before tool counts

List your five required tasks. For each one, check whether the connector exposes the necessary reports, filters and actions. A service limited to reads cannot publish a Tag Manager version. A service that supports writes may let you restrict them; inspect the actual permission model.

Also check support for date ranges, website timezones, pagination and error handling. Those affect whether an answer is reliable even when both products can retrieve a visits summary.

Test one reporting workflow

Start with website discovery, choose a property explicitly and compare two completed periods. Ask the assistant to show original figures, dates and metric definitions. Verify the answer in Matomo before evaluating a longer workflow.

Use the ChatGPT walkthrough or the Claude walkthrough for Savigny. MCP provides the connection; the assistant and the available report data determine the usefulness of the final explanation.