Comparisons
Bonnard vs Metabase: MCP server features compared
Choose Metabase when you want agents to build the dashboards and alerts your team works in, on software you can host yourself for free. Choose Bonnard when you do not want to host or run a BI tool, and want answers in Claude or ChatGPT on a free hosted plan.
On this page
With Metabase (metabase.com), the agent works through a Metabase instance: it reads your metrics and glossary there, and can save dashboards. With Bonnard (bonnard.ai), no BI tool sits between the agent and your database, and every query is read-only.
Five differences
What the agent knows about your data
Bonnard gives its context to the AI client the person already uses. explore_schema searches table names, descriptions, column comments, enum labels, metrics and business terms, and returns up to 10 tables. For each table the agent gets the keys, row count and sample rows from the database, plus what your team adds: descriptions, joins, metrics, business terms and the values of columns you mark as searchable (context). A metric is its calculation in SQL, with instructions. The agent reads it and writes the query. Bonnard does not rewrite the query to apply it.
Metabase also gives its context to the AI client. search finds questions, models, metrics, measures, segments, dashboards and tables, and returns each with its description. browse_data returns tables and fields with their types and sample values. get_content returns the definition of a metric, measure or segment, and glossary returns your team’s business terms (MCP tools). The glossary, measures and segments are in all plans. The Library and verified content are on Pro and Enterprise (Metabase pricing).
Metabase’s server instructions tell the agent to read the glossary before it answers a question about a business term, and to use execute_query unless the question needs SQL (server source). execute_query takes Metabase’s query language, so the agent can still write its own count or sum on a table. An admin can block raw SQL through MCP for the whole instance (Metabase 64). Where raw SQL is allowed, execute_sql needs native-query permission, and Metabase marks it as a tool that can overwrite or delete data.
How wrong answers get found and fixed
Bonnard runs a loop of four steps on how agents use your data. Metabase has the first step, on Pro and Enterprise, and leaves the other three to a person.
- ObserveAgents use your data
- DiagnoseBonnard finds what’s wrong
- ImproveYour data team answers, into the context
- VerifyEvals check it helped
Observe. Bonnard records each tool call with its SQL, status and error, from every AI client your people use.
Metabase has MCP analytics on Pro and Enterprise. A usage tab charts calls by client, tool and user. A tool calls tab lists each call with its tool, client, user, status, duration and error type, and shows error messages only when PII retention is on (MCP docs). Conversation transcripts, generated queries and thumbs-up or thumbs-down ratings are in AI usage auditing, which covers Metabot, Metabase’s own agent. That page states “MCP activity isn’t included in Usage auditing” (AI usage auditing).
Diagnose. Bonnard reads that record and finds what is wrong: definitions that differ between people, searches that found nothing, sessions that took many calls, repeated errors and tables that stopped updating. Where the data cannot settle a question, Bonnard asks your data team.
Last 30 days- Inconsistent definitionsQuestion
net revenue3 definitions, 8 people - Searches that found nothingData gap
"gift cards"6 people - Struggling work4 sessions, 9 calls eachContext gap
- ErrorsContext gap
undefined_column14 calls - Stale or misleading dataContext gap
orders_oldstopped in May
Metabase’s docs describe no suggestions generated from MCP activity. A person reads the analytics and decides what is wrong.
Improve. Your data team’s answer goes into a draft of the context.
On Metabase, a person changes the glossary, metrics or descriptions by hand. Questions, dashboards, metrics and models keep up to 15 versions, and segments and measures keep a revision history (history). On Pro and Enterprise, Remote sync stores the Library, glossary included, as YAML in git. A team reviews a change in a pull request, and a read-only production Metabase pulls it (Remote sync).
Verify. Evals run the same questions against the draft and the live version, and show which answers got better. A person then publishes the draft, and Bonnard keeps each version.
- Which products brought in the most revenue in August?Better
- What is our average order value?Better
- What share of orders were refunded?No clear difference
Metabase’s docs describe no evals of agent answers.
Two warnings also reach the agent as it works. A result over the row limit tells the agent not to sum or average it. On PostgreSQL, a join that could count rows twice gets a warning.
Metabase shows an admin which calls failed. Bonnard also finds what is wrong and tests the change before a person publishes it.
What you need first
Bonnard starts from your database. Add it as a data source through a read-only credential, then give people access. The agent reads the table comments, keys and enum values that the database declares, and the context your team adds in Bonnard.
Metabase starts from a Metabase instance, on Metabase Cloud or on your own servers, with your database connected to it. An admin sets the MCP server to on, and AI features must be on. Metabase needs no AI provider for MCP (MCP docs). The context is as good as the descriptions, metrics, segments and glossary terms your team keeps in Metabase.
Who can connect
A Bonnard workspace has unlimited users, your customers’ users included. Each customer’s query runs as that customer’s tenant (the value that marks its data in your database). A customer with no tenant gets no data. Isolation uses the database: row-level security on PostgreSQL, Redshift and SQL Server, or one dataset or database for each tenant on BigQuery and Azure SQL Database (tenancy).
On Metabase, each person signs in to your Metabase through OAuth 2.0, and the token carries that person’s Metabase permissions (MCP docs). The consent page asks for up to 5 permission scopes, and raw SQL is one of them. On Pro and Enterprise, admins choose which MCP tools each group can use (usage controls). Also on Pro and Enterprise, an agent in your app can call the MCP server for the person signed in to your app. Your backend exchanges a signed JWT for a one-hour token, and Metabase creates or updates an account for that person. Everyone who signs in to Metabase counts as a user (Metabase pricing), so ask Metabase how JWT accounts are billed.
How you pay
Bonnard bills for each call, with no seats or minimums. Free includes 3,000 calls a month, then USD 0.95 for each 100 calls once an admin turns on extra usage (pricing).
Metabase includes the MCP server in all plans. Open Source is free on your own servers, with unlimited users. On Metabase Cloud, Starter is USD 100 a month with 5 users, then USD 6 for each user a month. Pro is USD 575 a month with 10 users, then USD 12 for each user a month, in the cloud or on your own servers. Enterprise starts at USD 20,000 a year. Starter and Pro have a 14-day free trial (Metabase pricing). MCP calls do not count against Metabase’s AI connection, because the AI client brings the model (MCP docs). MCP analytics, per-group tool access, verified content and Remote sync need Pro or Enterprise.
Side by side
| Feature | Bonnard | Metabase |
|---|---|---|
| Who answers | The AI client the person uses, with Bonnard’s context | The AI client the person uses, with Metabase’s content. MCP calls do not use Metabot |
| Context the agent reads | Schema, descriptions, joins, metrics, business terms, searchable values | Tables, fields, sample values, descriptions, metrics, measures, segments, glossary terms |
| Context through MCP | explore_schema, for any AI client | search, browse_data, get_content and glossary, for any AI client |
| Metric definitions | The agent reads the SQL and writes the query | The agent reads the definition and builds the query. An admin can block raw SQL |
| Observe | Records each tool call, from any AI client | MCP analytics list each call, on Pro and Enterprise |
| Diagnose | Finds what is wrong in the activity record | By hand |
| Improve | The data team’s answer goes into a draft | By hand. Up to 15 versions of an item, and pull requests through Remote sync on Pro and Enterprise |
| Verify | Evals: the draft against the live version | None |
| Run a query | sql_query, one read-only statement on your database | execute_query, Metabase’s query language, or execute_sql, SQL that can change data |
| Charts and dashboards | render_chart and render_dashboard, in the AI client | visualize_query, in the AI client. dashboard_write saves dashboards in Metabase |
| Who can connect | Unlimited users, your customers’ users included | People who sign in to your Metabase. On Pro and Enterprise, also your app’s users through a JWT exchange |
| Needed first | A data source: your database, or the sample data | A Metabase instance with your database connected |
| Databases and warehouses | 7, plus PostHog | 18 with official drivers, plus community drivers |
| Tools for the agent | Up to 17 | 23, across 5 permission scopes |
| How you pay | For each call, no seats | Free on Open Source. A monthly fee with 5 or 10 users plus each further user on Starter and Pro. From USD 20,000 a year on Enterprise |
| Free plan | 3,000 calls a month | Open Source, on your own servers, unlimited users |
When Metabase is stronger
- You want to host BI yourself for free: Open Source includes the MCP server and has unlimited users (pricing).
- You keep context in git: on Pro and Enterprise, Remote sync stores metrics, segments, measures and the glossary as YAML, and a change goes through a pull request (Remote sync).
- You want admins to limit what agents can do: Metabase has 5 permission scopes on the consent page, per-group tool access on Pro and Enterprise, and a block on raw SQL through MCP (usage controls).
- You want agents to build Metabase content:
dashboard_write,metric_write,alert_writeandsubscription_writesave dashboards, metrics, alerts and scheduled deliveries by email or Slack (MCP tools). - You want drill-through on a chart in the chat:
visualize_querycharts zoom, sort and drill, and each has an Explore in Metabase button (MCP docs). - You need more database engines: Metabase has official drivers for 18, plus community drivers (databases).
Add each server to your AI client
Bonnard. Create a workspace, add Bonnard to your AI client, then add a data source (quickstart). Use the install links for Claude, Cursor and ChatGPT in the console at app.bonnard.ai, or add this URL.
https://app.bonnard.ai/mcpMetabase. An admin sets the MCP server toggle to on in Admin > AI > MCP. On the same page, the admin sets the toggle to on for each client that must show inline charts. In Claude on the web or Claude Desktop, add Metabase from the connector directory and enter your Metabase’s MCP URL. Other clients use this URL.
https://<your-metabase>/api/metabase-mcpIn Claude Code, add the server with this command.
claude mcp add --transport http metabase https://<your-metabase>/api/metabase-mcpVerified 16 Nov 2026 against Metabase’s MCP docs, MCP tools docs, pricing, usage controls, AI usage auditing, history, Remote sync, content verification, Metabase 64 release notes, Metabase 64 changelog, Metabase 65 release notes, Metabase 65 changelog, databases and server instructions.
Get started
Start scaling accurate data today
Free for your first 3,000 calls. No seat limits.
