Metabase is not treating AI as a dashboard garnish. With Metabase 60, it is turning the BI layer into an agent-facing surface: MCP server, Agent API, Metabot in Slack, file-based development, and AI features available across plans. ToolVitals sees the same product as unusually healthy for a mature BI project: 47,476 GitHub stars, 30 GitHub releases in 90 days, 3 release events in 30 days, a 97 shipping score, and a 98 ToolVitals score.

That combination matters. A lot of analytics tools are trying to bolt natural language onto charts. Metabase is making the semantic layer callable by agents while keeping the permission model in the loop. Its homepage now leads with “open source analytics that answers back” and describes natural-language querying over metrics and permissions. The Metabase 60 release goes further, positioning AI as something grounded in curated datasets, reusable metrics, defined segments, and governed access.

The strongest signal is not just the MCP server. It is the way Metabase is wrapping AI into product, developer workflow, and community at once. The April hackathon asked users to build with the MCP server, Agent API, file-based development, or Metabot in Slack. The winners then used those pieces for agent-driven chess dashboards and personal fitness advice written back into Metabase. That is product-market exploration in public, not a quiet feature flag.

Metabase also looks like a team dogfooding agent tooling under real codebase pressure. Its engineering post about ten custom subagents describes a 500K-line Clojure backend with domain-specific agents for query processing, permissions, drivers, notifications, and more. A separate security post says public source projects are seeing more LLM-powered vulnerability scanning. That context makes the Metabase 60 AI push feel less like a press release and more like a company adapting its own engineering habits to its product.

Open core, not generic open source

ToolVitals classifies Metabase as open core. That distinction matters. Metabase has an open-source core and paid commercial editions, and hosted cloud pricing is tracked separately.

First-party pages use open-source language, and the Metabase 60 release says AI features are available across plans. The GitHub release pages for 59.8 and 60.2 also list separate Open Source and Enterprise Docker images and JAR downloads. For ToolVitals, the conservative phrasing is simple: Metabase is an open-core BI tool with a public repo, a self-hosted open-source path, and commercial editions.

What ToolVitals cannot infer

ToolVitals can see public activity signals. It can say Metabase has 47,476 GitHub stars, 30 releases in 90 days, 3 release events in 30 days, a 98 health score, and 100 data confidence. It can also see that the website is alive and that release and RSS signals are fresh.

It cannot tell you whether Metabot answers your finance team’s questions correctly. It cannot measure dashboard quality, support response time, customer satisfaction, query performance on your warehouse, or whether the Agent API fits your security review. The public signals say Metabase is maintained and shipping. They do not prove it is the right BI layer for your org.

The comparison is surprisingly close to infrastructure tools

Metabase is nearly the same size as ClickHouse by stars in this snapshot: 47,476 for Metabase versus 47,666 for ClickHouse. ClickHouse is moving faster on short-term release events, with 32 in 30 days and a 100 shipping score, compared with Metabase at 3 release events in 30 days and a 97 shipping score. That is a fair gap, but it is not a warning sign for Metabase. BI products do not need database-style release volume to look alive.

n8n is much hotter on ToolVitals, with 190,094 stars, 21 release events in 30 days, a 100 shipping score, and a 240.0 hot score. It is also fair-code, not OSI-approved open source. Metabase sits in a different lane: fewer short-term release events than n8n, but a very high health score and an open-core model built around BI and embedded analytics rather than workflow automation.

Recommendation

If your team already trusts Metabase for dashboards and wants AI agents to query governed metrics instead of raw warehouse tables, evaluate Metabase 60 now. The reason is specific: the MCP server and Agent API put agents behind the same semantic and permission layer your analysts already maintain.

If you need pure OSS licensing, do not blur the line. ToolVitals classifies Metabase as open core, not OSI-approved OSS. If that passes your policy, the public signals are strong enough to justify a serious trial. If it does not, compare against OSI-approved alternatives before you build workflows around it.

Sources