Metabase

What it is

Metabase is an open-source business intelligence and embedded analytics tool aimed at letting non-technical people ask questions of a database through a point-and-click query builder, with SQL for power users. Its repo describes it as “the easy-to-use open source Business Intelligence and Embedded Analytics tool that lets everyone work with data” (GitHub README, fetched 2026-10-05). Part of the open-source BI group: see Open-source BI comparison.

Maker and history

Made by Metabase (company). The GitHub repo (created 2015-02-02, about 49.6k stars on 2026-10-07) holds both the open-source edition, released under the AGPL, and the commercial editions, released under the Metabase Commercial Software License (README). Latest stable release on 2026-10-07: v0.64.1 (0.64.0-beta was published 2026-09-24). Company founding date, funding and leadership: not verified (see Open items).

Products and editions

• Open Source (AGPL, self-host, free).
• Pricing page (2026-10-07) names Open Source, Starter, Pro and Enterprise; Starter and Pro list a 14-day free trial; Cloud (managed) and self-hosted are both offered; AI, transforms and storage are optional usage-based extras. Pricing: see the vendor pricing page, https://www.metabase.com/pricing (not fetched, no amounts recorded).

Data connectivity and modelling

Connects to SQL databases such as MySQL and PostgreSQL and many more. Modelling is lightweight: saved questions, models, metrics and table metadata in the UI, plus native SQL. It is not a code-defined semantic layer like Lightdash.

Deployment

Self-hosted (Docker/JAR) or Metabase Cloud; embedding in other apps is a core use case.

AI features as of 2026-10-05

Source: Metabase docs “AI overview”, “Metabot” and “MCP” pages.

• Metabot: built-in AI agent for data questions, chart generation from natural language, SQL generation and SQL error fixing, explaining charts, creating documents, generating code for transforms, and answering questions in Slack. It only sees what the asking user can see. Admins get usage controls (token limits, tool restrictions). Docs: “available on all plans”. It needs either the Metabase AI service (Cloud add-on) or your own third-party model-provider credentials (docs AI overview, 2026-10-07).
• MCP server: lets external AI clients (Claude, Codex named in docs) connect to a Metabase instance via OAuth 2.0 using Metabase’s embedded OAuth server; responses are scoped to the user’s permissions. Tools: interactive (visualisations), read-only (query, search tables/metrics) and write (create dashboards, questions, collections). Admin must enable it (Admin > AI > MCP). Docs do not label it beta or experimental and say it is “available on all plans”; the page mentions clients connected before Metabase 0.64 must sign in again after upgrading, so the server predates 0.64 (introduction version not stated). Scopes include agent:content:read/write, agent:query:run, agent:sql:run, agent:delivery:write; MCP usage analytics need Pro/Enterprise.
• Agent-driven development: coding agents plus the Metabase CLI create content as version-controlled YAML files. Pro, Enterprise, Cloud only.
• Metabase AI Service: proprietary, Cloud add-on only; self-hosted instances supply their own model-provider credentials.
• Caveat: docs warn feedback forms may contain sensitive conversation data; Metabot uses either the Cloud AI service or your own provider (specific provider/model not named).
• Limits listed: no SQL parameters/variables in generated queries, no chart formatting changes, cannot manage existing alerts.

Strengths and weaknesses (opinion, from docs and the 2026 comparisons listed in Sources)

• Strengths: point-and-click query builder aimed at non-technical users (README); the docs say Metabot and MCP are “available on all plans”, though Metabot needs the Cloud AI add-on or your own provider credentials.
• Weaknesses: less code-first governance than Lightdash/Evidence; AGPL may matter for embedding (check with counsel).

Self-learning

• Metabase Learn, https://www.metabase.com/learn - Metabase, beginner to advanced, free, guides of 1-15 min (opened 2026-10-05): Getting Started /learn/metabase-basics/getting-started, Learn SQL /learn/sql, Database fundamentals /learn/grow-your-data-skills/data-fundamentals, Administration guide, Querying and dashboards.
• Metabase docs, AI section: https://www.metabase.com/docs/latest/ai/overview - free.
• Metabase MCP docs: https://www.metabase.com/docs/latest/ai/mcp - free.
• No vault video summaries found in yt-base for this tool.

Sources

• https://github.com/metabase/metabase (2026-10-05)
• https://www.metabase.com/docs/latest/ai/overview, /ai/metabot, /ai/mcp (2026-10-05)
• https://www.metabase.com/learn (2026-10-05)
• https://www.metabase.com/pricing (2026-10-07: plan names, trials; amounts deliberately not recorded)
• https://www.metabase.com/docs/latest/ai/overview, /ai/metabot, /ai/mcp (re-fetched 2026-10-07)
• https://api.github.com/repos/metabase/metabase and /releases, README (2026-10-07)

Open items

• Company founding date, founders, funding: not verified (Wikipedia page returned 404).
• Whether MCP/Metabot are GA or beta: docs carry no beta label, but no explicit GA statement either.
• Exact version that introduced the MCP server.