Bonnard
On this page

With Looker (cloud.google.com/looker), the agent picks fields from one Explore and Looker writes the SQL. With Bonnard (bonnard.ai), the agent writes the SQL, from your schema and the context your team adds.

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.

Looker gives its LookML model to the AI client too. The Looker-managed MCP server has been generally available since 26 Oct 2026 for Looker-hosted instances of Looker (Google Cloud core) and Looker (original) (release notes). Google lists 30 tools for it (tool list). get_models and get_explores list the models and their Explores. get_dimensions and get_measures list the fields of an Explore. In MCP Toolbox for Databases, the open-source server with the same tool names, each field comes with its label, description, tags, synonyms and suggested filter values (get_dimensions reference).

Looker enforces the model. query takes a model, an Explore, fields, filters and sorts, and Looker writes the SQL from the LookML (LookML). No tool in Google’s list takes SQL from the agent. query_sql returns the SQL that Looker generates. The agent can answer only from fields that a developer has modelled.

Looker has a second set of context for its own agent. A Conversational Analytics data agent holds instructions and verified queries over up to five Explores (data agents). None of the 30 tools reads a data agent’s instructions or verified queries.

How wrong answers get found and fixed

Bonnard runs a loop of four steps on how agents use your data. Looker has the first step for MCP, a log of each action, and leaves the other three to your team.

01020304 OBSERVEAgents useyour dataDIAGNOSEBonnard findswhat’s wrongIMPROVEYour datateam answers,into the contextVERIFYEvals check it helped
01020304 OBSERVEDIAGNOSEIMPROVEVERIFY
  1. ObserveAgents use your data
  2. DiagnoseBonnard finds what’s wrong
  3. ImproveYour data team answers, into the context
  4. VerifyEvals check it helped

Observe. Bonnard records each tool call with its SQL, status and error, from every AI client your people use.

Looker logs each MCP action in System Activity, in the History and Event Attribute Explores. Looker (Google Cloud core) also logs it in Cloud Audit Logs (Looker MCP docs).

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.

On Looker, an admin reads those logs by hand. Looker has no feature that turns MCP activity into suggested model changes. Its review features cover Conversational Analytics, its own agent. Users mark a response good or bad (Conversational Analytics). The Responses & feedback tab of the Conversational Analytics dashboard shows answer success, ratings and each rated question, in Preview, for users who consent (System Activity).

Improve. Your data team’s answer goes into a draft of the context.

On Looker, a developer edits the LookML by hand. Each LookML project is a git repository, each developer works on a branch, and a project can require a pull request (version control). For Conversational Analytics, an agent editor adds a verified query, a question paired with an Explore URL, to steer later answers. Google recommends 30 to 50 for each agent.

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.

Looker has no evals of agent answers. LookML data tests must pass before a deploy, by default on new projects (version control). The tests check the model’s data, not an agent’s 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.

On Looker, your team finds the gaps in the logs and writes each fix in LookML. Bonnard reads the activity record and proposes the fix.

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.

Looker starts from a Looker-hosted instance with a LookML project on a database connection. A table with no LookML view and Explore is out of the agent’s reach through query. A new production instance of Looker (Google Cloud core) needs an annual contract and quota from Google sales (create an instance). Customer-hosted instances cannot use the managed server (Looker MCP docs). All tools are off until a Looker admin selects them on the MCP page of the Admin panel (MCP settings).

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 Looker, each person signs in with their own Looker account through OAuth 2.1. The agent gets the Looker roles and content access of that user. A Looker admin enables each AI client in the Application Library, or registers a custom OAuth client if it is not listed (Looker MCP docs). Dynamic client registration is not supported.

Looker’s MCP server has no route for your customers’ users. Looker serves external users through signed embedding, outside MCP. On Looker (Google Cloud core), signed embedding needs the Embed edition (signed embedding).

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).

Looker (Google Cloud core) has three platform editions on an annual commitment: Standard, Enterprise and Embed. Each includes 10 Standard Users and 2 Developer Users. Other users need a Developer User, Standard User or Viewer User licence. Google publishes no edition or licence prices, and each edition’s cost reads “Call sales” (Looker pricing). Standard User and Viewer User licences do not include the Looker API interfaces, so ask Google which licences can use the MCP server.

The managed MCP server has no extra cost. Its tool calls count toward the instance’s API quotas (Looker MCP docs). Standard includes up to 1,000 query-based API calls a month, Enterprise 100,000 and Embed 500,000. Conversational Analytics is billed apart, in data tokens. From 1 Feb 2027, use over the monthly allocation costs USD 3 for each million input tokens and USD 20 for each million output tokens.

Side by side

Bonnard and Looker MCP servers, verified 16 Nov 2026
FeatureBonnardLooker
Who answersThe AI client the person uses, with Bonnard’s contextThe AI client the person uses, with the LookML model
Context the agent readsSchema, descriptions, joins, metrics, business terms, searchable valuesModels, Explores, dimensions and measures, with labels, descriptions, tags and synonyms
Context through MCPexplore_schema, for any AI clientget_models, get_explores, get_dimensions and get_measures, for any AI client an admin enables
Metric definitionsThe agent reads the SQL and writes the queryEnforced. query takes LookML fields and Looker writes the SQL
ObserveRecords each tool call, from any AI clientSystem Activity and Cloud Audit Logs
DiagnoseFinds what is wrong in the activity recordBy hand. No suggestions from MCP activity
ImproveThe data team’s answer goes into a draftBy hand, in LookML through git
VerifyEvals: the draft against the live versionNone for agent answers. LookML data tests check the model
Run a querysql_query, one read-only statement on your databasequery, fields from one Explore. No SQL from the agent
Charts and dashboardsrender_chart and render_dashboard, in the AI clientquery_url, make_look and make_dashboard return links to Looker
Who can connectUnlimited users, your customers’ users includedPeople with a Looker account on the instance
Needed firstA data source: your database, or the sample dataA Looker-hosted instance with a LookML project
Databases and warehouses7, plus PostHog39 SQL dialect options on Looker (Google Cloud core)
Tools for the agentUp to 1730, all off until an admin selects them
How you payFor each call, no seatsA platform edition and user licences, through sales
Free plan3,000 calls a monthNone. A trial edition lasts 90 days

The dialect count is from Google’s dialect list, and the trial length from the instance docs.

When Looker is stronger

  • You want metric definitions enforced: every query call uses LookML measures, and the agent cannot write its own SQL. Bonnard gives the agent the metric’s SQL and does not rewrite the query.
  • You already have a LookML model: its Explores, descriptions and synonyms reach the AI client with no second copy to keep.
  • You keep the model in git: branches, pull requests and data tests gate each change (version control).
  • You want agents to write the model: in Development Mode, create_project_file and update_project_file change LookML files (tool list).
  • You want an agent inside the BI tool: Conversational Analytics answers in Looker, on dashboards and in embedded content, with verified queries and user ratings (Conversational Analytics).
  • You need more warehouses: Looker (Google Cloud core) supports 39 SQL dialect options against Bonnard’s 7 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/mcp

Looker. A Looker admin selects the tools on the MCP page of the Admin panel, then enables Claude Desktop in the Application Library on Manage OAuth clients and copies its Client ID. In Claude Desktop, select Settings > Connectors > Add custom connector. Enter this URL, and enter the Client ID under Advanced settings. Leave the client secret blank, then restart Claude Desktop.

https://<your-looker-instance>/mcp

For a customer-hosted instance, run MCP Toolbox for Databases with Looker API credentials or an OAuth app instead. Google does not cover it under its support terms.

Verified 16 Nov 2026 against Google’s Looker MCP docs, release notes, MCP settings, tool list, pricing, LookML, version control, Conversational Analytics, data agents, System Activity, signed embedding, instance creation, dialect list and MCP Toolbox.

Get started

Start scaling accurate data today

Free for your first 3,000 calls. No seat limits.

Keep reading