The MCP tools a Bonnard app offers

An MCP tool is one operation that an app offers to an agent, the model in an AI client that acts for a person. The agent calls tools to find tables, run SQL and draw charts. In the console at app.bonnard.ai, the app’s Tools tab sets which tools the agent gets. Two behaviours add no tool: Correct a wrong name and Track call intent.

Verified 15 Oct 2026.

Each setting has a switch on the Tools tab, unless the table says otherwise.

Change a setting

You need the Admin role in your Bonnard workspace.

1. In the console, select Apps, then the app, or Open app details in the setup steps.
2. Select the Tools tab.
3. Select the switch next to the setting. The change saves immediately.

A row with no switch shows included. When a setting is off, its row adds “Off: hidden from agents.” and the app does not offer its tools.

The switch for call intent is on the Settings tab, in the Track call intent entry.

Bonnard applies a change to a setting from the next request to the app. The change applies to the staging URL and the production URL. See Publish and staging.

Docs

explore_docs gives the agent reference pages to read before it builds a chart or a dashboard. The pages cover chart types and their options, KPI tiles and dashboard params. It shows only the pages for the tools that the app offers.

The Docs row has no switch and shows included. The app offers explore_docs only while Charts or Personal dashboards is on.

Explore schema

explore_schema gives the agent an overview of the database and finds tables by a concept. For one table, it returns the columns, their types, the row count and sample rows.

The tool answers only with the tables that the app shows. Tables that you hide on the app’s Data tab are not in its answers. See Table curation.

Always included. Explore schema and SQL query have no switch, because an app without them has nothing to do.

SQL query

sql_query runs one SQL query against the app’s database and returns the rows. Table curation is not access control: a query that names a hidden table runs.

Always included.

Charts

render_chart draws a chart from a SQL query, or a carousel of up to 10 charts. fetch_view_data is for the chart widget, which calls it to load a chart’s data.

On by default. When Charts is off, the agent cannot draw a chart with render_chart. save_chart then cannot save a new chart, because it saves only what render_chart drew. If Personal dashboards is off too, the app does not offer Docs.

Export data

create_export_file makes a file from a result or a chart that the agent has: a CSV, a gzip-compressed CSV or a PNG image. The file comes from the result that Bonnard captured when the query ran, so the query does not run again.

While Export data is on, sql_query also takes two arguments, capture_only and max_rows. With them, the agent can capture more rows for a file than the standard limit of 5,000 rows.

On by default. When it is off, the agent cannot make files, sql_query loses the two arguments and the app does not offer save_chart.

Personal dashboards

From the AI client, each person who uses the app can keep their own dashboards. save_dashboard saves a new dashboard or changes a saved one, and delete_dashboard deletes one.

The app also offers the Published dashboards tools and Docs, so that the agent can show and draw the saved dashboards. This applies even when Published dashboards is off. In that case, the tools offer only personal dashboards.

On by default for a new app. An app made before 14 Oct 2026 has it off, unless an admin changed it. When it is off, people cannot save dashboards from the AI client.

Personal charts

save_chart saves a chart that the agent drew, explore_charts shows the saved charts, and delete_chart deletes one. Each person gets only their own charts.

save_chart saves from a captured result, so it needs Export data. While Export data is off, the app does not offer save_chart.

On by default. When it is off, the app offers none of the three tools.

Published dashboards

explore_dashboards shows the dashboards that the app offers, and render_dashboard draws one. The dashboard widget calls search_param_values to fill its filter menus, and fetch_view_data to load data.

Off by default. When it is on, the row shows the published dashboards of the app’s database, each with a checkbox:

1. Select the checkbox of each dashboard that the app offers. Each change saves immediately.
2. To add or change dashboards, select Manage dashboards →. It opens the Dashboards page.

All checkboxes start clear. Until you select one, the app offers none of these tools, unless Personal dashboards is on. If the database has no published dashboard, the row shows “No dashboards yet for”, then the database name, and a Create a dashboard button.

The production URL serves the published version of each selected dashboard. The staging URL serves its current draft.

Correct a wrong name

This behaviour covers table names only. When an agent names a table that does not exist, Bonnard adds the closest real table names to the error. This applies to explore_schema with a table name, and to sql_query.

The same error also names up to 25 of the app’s tables. Hidden tables are never in the suggestions. A wrong column name gets no suggestions: the error tells the agent to read the table’s columns with explore_schema.

On by default. The console has no control for it.

Track call intent

While this behaviour is on, each tool takes an optional intent argument of up to 300 characters: why the agent makes the call. Bonnard asks the agent for one sentence of 15 to 25 words, with no names, emails, IDs or row values. The intent never changes the result. Bonnard records it with the call.

On by default. To change it, select the switch on the Call intent card of the app’s Settings tab. When it is off, the tools have no intent argument and Bonnard records no intent.

The console does not show the recorded intents.