Bonnard

Works in any MCP-compatible app

  • Claude
  • ChatGPT
  • Gemini
  • Cursor
  • Langdock
  • Goose
  • Mistral
  • Archestra
  • Glean
  • Copilot

We tested Bonnard against some of the most popular BI tools.

+26% accuracy

42% faster

13k tokens

20 business questions, 3 tries each, on a dbt-built warehouse with metric definitions loaded.

Correct answers
vs the average of six BI MCP servers, same 20 questions
Time per answer
22s vs 38s for the next fastest
Tool output per answer
Next lowest: 25k
Method on request
For agents

Fastest to the answer.
Lightest on tokens.

Against six other BI products we tested, on a dbt-built warehouse with metric definitions loaded.

Answers from your own definitions.

Without Bonnard

An agent that sees only table and column names guesses what each metric means. A wrong guess still returns a number that looks right.

With Bonnard

Bonnard gives the agent the full definition of each metric, from the descriptions already in your warehouse, and never cuts it short.

Correct answers+26% accuracyvs the average of six BI MCP servers, same 20 questions

Half the calls. Half the reading.

Without Bonnard

Most of an agent's time goes on finding tables and reading results. A result too large for the chat is saved to a file, and the agent reads it back in full.

With Bonnard

Bonnard lists your exact tables in one call and sends short previews, keeping the rest on the server. Against the next fastest product, the agent makes half the calls and reads half the text.

Time per answer42% faster22s vs 38s for the next fastest

Only the rows it needs.

Without Bonnard

A query that returns thousands of rows can land in the chat in full, and you pay for every one in tokens.

With Bonnard

Bonnard sends a short preview and keeps the rest on the server, ready to save or export without running the query again.

Tool output per answer13k tokensNext lowest: 25k
Trust

You decide who connects and which rows they read

Governance

Sharing

  • Hosted and shared. Share any chart or dashboard by link, with your team or customers.
  • Edit in Bonnard. Edit saved dashboards in Bonnard’s web app.

Access

  • An audience per app. Anyone outside it is refused.
  • Every call, logged. Who used it, when, which tools, and what failed, scoped per app and per customer.

Publishing

  • Staging for drafts. Each app has its own staging endpoint.
  • Production on publish. The production endpoint answers once you publish.

Security

Sign-in

  • OAuth 2.1, PKCE mandatory. Tokens are audience-bound per server.
  • Bring your own IdP. OAuth, API keys or trusted JWTs. SSO supported.

Data

  • AES-256-GCM at rest. Credentials encrypted. API keys stored only as hashes.
  • Read-only. SELECT statements only. Connects over SSL from fixed IPs.

Tenancy

  • Set per database. Row-level security, or a schema, database or dataset per tenant.
  • Tenant from the sign-in. An IdP sign-in maps to the right tenant. Scale and Enterprise plans.
Setup

Four steps,
no engineering

  1. Connect your data

    Databases, warehouses and PostHog, connected read-only.

    • PostgreSQL
    • BigQuery
    • Redshift
    • ClickHouse
    • MotherDuck
    • MySQL
    • MariaDB
    • SQL Server
    • Azure SQL
    • PostHog
  2. Create an app

    Choose the tables it can see and the tools it offers: charts, dashboards, exports.

  3. Choose the audience

    Your team, all your customers, or the ones you name.

  4. Share the link

    Users connect from the AI client they already use.

Pricing

Unlimited users, usage-based pricing

Start free with 3,000 calls a month and every BI feature.