OpenAnalytics ships v0.6.0, a self-hosted analytics stack an agent can query over MCP

OpenAnalytics released v0.6.0 on 5 September 2026. It is a web analytics platform you run yourself, licensed AGPL-3.0, and the release adds an MCP server so a coding agent can read the same numbers the dashboard shows. The GitHub API reported 401 stars and 38 forks when checked on 6 September, against a repository created on 11 August.

The project describes itself as a Plausible and Google Analytics alternative. That places it in a category the incumbents have no reason to advertise: the data sits on hardware you control, and the code that reads it is one you can change.

What v0.6.0 adds

The release notes list filters that narrow the overview to matching sessions when you click a row in Sources or Locations, entry and exit counts with bounce rate per page, and click ID attribution for gclid, fbclid and msclkid so a visit is credited to the ad network even when the referrer is empty. Polar joins Stripe as a second revenue provider. There is an importer for Umami data, and MCP support with hints and native-app redirects.

The MCP server is the part that changes a working pattern. It puts traffic, funnels and revenue behind the same protocol an agent already uses for other tools, so the question "which pages lost traffic this week" can be answered by something that also has the repository open.

The privacy model is enforced in the code, not promised

No cookies and no fingerprinting. Visitor identity is a salted hash that rotates nightly, raw IP addresses are never stored, and Do Not Track and Global Privacy Control are honoured at the collector before anything is written. City-level geolocation is opt-in per site and resolves against a database on your own disk, so no lookup leaves the host.

Four of the architecture rules are checked by CI rather than left as convention. The dashboard may import only the OpenAPI contracts package. ClickHouse is readable only through a gateway that verifies Ed25519-signed query envelopes. The browser tracker has a byte budget that fails the build when exceeded. A CI job asserts against a real database that the schema contains no billing tables.

What it costs to run

The stated requirements are a Linux host with Docker, four DNS records, about 4 GB of RAM and 25 GB of free disk. Ten images are published to the GitHub container registry, so a fresh host pulls rather than builds. Images are amd64; on arm64 you build the ten yourself, which the project puts at about ten minutes.

Two warnings are worth reading before adoption. Migrations do not run backwards, so the way back from a bad upgrade is a restore, and a restore discards everything that arrived after the upgrade. The four DNS names must resolve before first boot, because certificates are issued then and the failure surfaces half an hour later nowhere near its cause.

The stack underneath is Postgres, ClickHouse and two Valkey instances, which is more moving parts than a single-binary analytics tool. That is the trade for funnels and revenue reporting on your own hardware. AGPL-3.0 also carries an obligation people miss: run a modified copy as a network service and you owe your users the modified source. Also worth knowing is that the name is not part of the licence grant, so a self-hosted instance runs the software without the brand.

Source

OpenLabs-so/openanalytics, GitHub. Release v0.6.0 published 5 September 2026.