Metabase is not just shipping BI maintenance releases. The recent signal is sharper: Metabase is pushing its analytics product toward agent-facing workflows, with 10 release events in 30 days, 30 GitHub releases in 90 days, a 100 shipping score, and a 94 ToolVitals score.

The product positioning matches that shift. Metabase now describes itself as analytics that can answer back, with natural-language querying and AI-powered analytics built on top of existing metrics and permissions. Metabase 60 adds an official MCP server, an Agent API, Metabot in Slack, file-based content editing, semantic search, and other AI-adjacent features.

That is the interesting part. Metabase already had the installed base and BI surface area. The new bet is that analytics should become programmable by agents without bypassing the semantic layer.

The agent layer is the story

The April hackathon announcement is blunt about the direction. Metabase framed the MCP server, Agent API, file-based development, and Metabot in Slack as tools for plugging analytics into agents and workflows. The winners post then showed those features used in odd but concrete ways: a chess dashboard run through MCP and YAML-backed Metabase cards, and a nightly fitness-advice workflow that writes back into a chart.

That does not prove broad adoption. It does show that Metabase is trying to make its semantic layer useful outside the dashboard UI. That is a smart bet. If agents are going to touch business data, permissions, metrics, and data definitions matter more than a shiny chat box.

The engineering blog tells the same story from the inside. Metabase describes a 500K-line Clojure backend and explains how its team built ten Claude Code subagents around domain boundaries like query processing, permissions, and database drivers. Another post goes deep on cutting clojure-lsp startup time in half and memory allocation by two thirds. This is not random content marketing. It is a team showing how it is adapting a large BI codebase to AI-assisted development.

What ToolVitals can and cannot infer

ToolVitals can say Metabase is active. The payload shows 47,289 GitHub stars, 30 releases in 90 days, 10 release events in 30 days, a 100 health score, and 100 data confidence. The GitHub repository also shows recent activity and public release tags for multiple maintained lines.

ToolVitals cannot say the new AI features work well in production. It does not measure answer quality, permission edge cases, dashboard correctness, customer satisfaction, revenue, support burden, or whether teams actually trust AI-generated analysis. Those are product questions, not repository-health questions.

License language also needs precision. ToolVitals classifies Metabase as open core, with an open-source core and paid commercial editions. That is different from calling the whole tool OSI-approved open source.

How it compares

Among related tools, Metabase is slower than the automation outlier n8n on raw recent events: n8n has 50 release events in 30 days versus Metabase’s 10. But n8n is fair-code, not OSI-approved, and it sits in automation rather than data analytics.

Inside the data category, Metabase looks heavier and more mature than Jitsu on audience size, with 47,289 stars versus Jitsu’s 4,760. Jitsu has 30 release events in 30 days, compared with Metabase’s 10, so the signal is not simply “more releases equals stronger product.” Metabase’s signal is breadth: a mature BI product, multiple maintained release lines, and a visible push into AI-facing analytics.

CloudQuery is another useful contrast. It has 6,405 stars and 33 release events in 30 days, with an OSI-approved MPL-2.0 signal in the payload. Metabase has fewer recent release events, but a much larger GitHub audience and a clearer current narrative around governed AI access to business data.

Recommendation

If your team already cares about governed BI and wants agents to query data without bypassing permissions, evaluate Metabase now. The specific reason is not the 47,289 stars. It is the combination of a 100 shipping score, Metabase 60’s MCP and Agent API push, and an open-core model that lets you test the core product before paying for commercial editions.

If you only need an automation graph, compare it against n8n. If you need data movement or infrastructure inventory, compare it against Jitsu or CloudQuery. But if the job is analytics plus permissions plus agent access, Metabase is the one in this payload making the clearest move.

Sources