Bonnard
On this page

With Lightdash (lightdash.com), the agent picks metrics and dimensions, and Lightdash compiles the SQL from your YAML. With Bonnard (bonnard.ai), the agent reads each metric’s SQL and writes the query itself.

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.

Lightdash also gives its context to the AI client. The context is the semantic layer: tables, metrics and dimensions that your team defines in YAML, in a dbt project or a native Lightdash project (semantic layer). Each field can carry a description and an ai_hint, a note that only agents read (descriptions). Through MCP, list_explores lists the data models, grep_fields searches field names, labels, descriptions, hints and tags, and get_metadata returns the joins, required filters and full hints for the fields the agent picks (tool schema). list_verified_content lists the charts and dashboards that an admin marked as verified (Lightdash MCP docs).

run_metric_query takes an explore, metrics, dimensions and filters, and Lightdash compiles the SQL. The agent does not write the metric’s calculation, so two people get the same definition. The server’s instructions tell the agent to use run_metric_query for standard analysis and to keep run_sql for questions the explores do not model (instructions). run_sql shows only for users with the permission to manage the SQL Runner, which Developers and Admins have by default. For everyone else, the semantic layer is the only way to query. run_sql skips user attributes and row filters (user attributes). Lightdash’s FAQ says queries through MCP are read-only, and its docs tell you to give the warehouse connection read-only credentials, because a connection with write permissions could let an agent change data.

route_agent picks one of your Lightdash AI agents for the question. Calls that pass that agent’s ID are limited to its explores, and the AI client gets the agent’s instructions and verified answers (agent context).

How wrong answers get found and fixed

Bonnard runs a loop of four steps on how agents use your data. Lightdash records MCP calls, and documents the other three steps for its own AI agents, in the Lightdash app and in Slack.

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.

For calls from an outside AI client, Lightdash keeps a record. Settings > MCP > Analytics shows each tool call with its user, status and duration, for 90 days (visibility). For its own agents, admins read each conversation with its question, answer and query under Settings > Ask AI > Threads, and users rate answers with a thumbs-up or thumbs-down.

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.

Lightdash’s Issues, in beta and off by default, scans each agent turn, groups likely wrong answers by root cause and files them on a board.

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

On Lightdash, changes to context go through git. An issue can open a pull request that changes the semantic layer or adds an entry to the project context file. Metrics, dimensions, hints and the project context file are YAML in your repository. Agents, with their instructions and evaluation suites, can be downloaded as YAML and reviewed in pull requests (agents as code). Lightdash states that “Feedback alone does not rewrite the agent’s instructions” (effective analytics).

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.

Lightdash’s evaluation suites run a set of questions against an agent. An LLM grades each answer against the expected response you wrote, and you can add a thumbs-down answer to a suite. A preview project copies production, agents included, for each pull request.

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.

Lightdash’s Issues and evaluation suites are documented for its own agents’ conversations, so ask Lightdash whether they cover MCP calls. Bonnard’s loop reads the calls that Claude, ChatGPT or Cursor make.

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.

Lightdash starts from a project: a warehouse connection and a dbt or native Lightdash YAML project (connect a project). The YAML files describe your tables as explores, with their metrics and dimensions. A project can also start from the warehouse connection alone. It has no explores until you deploy YAML models, so grep_fields and run_metric_query have nothing to work on, and only SQL runs. A self-hosted instance also needs an Enterprise licence key and MCP_ENABLED=true (self-hosted MCP).

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 Lightdash, each person signs in with their own Lightdash account through OAuth, or a script sends a personal access token (Lightdash MCP docs). Metric queries run with that user’s permissions and row filters. A custom agent that serves many end users can use one org admin’s token. On each request, its X-Lightdash-User-Attributes header sets the user attributes, and so the row filters, for one end user. If the token’s user has a value for an attribute, the header can only narrow that value. run_sql skips those filters, so remove the SQL Runner permission from that token’s role, or enforce access in the warehouse.

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).

Lightdash’s MCP server is available to all Lightdash Cloud users (Lightdash MCP docs). Cloud Pro starts at USD 3,000 a month, with unlimited users and no fee for each seat. AI features, MCP included, are billed by usage on top, at a rate Lightdash does not publish (Lightdash pricing). Enterprise has no published price. The Open Source plan is self-hosted, and self-hosted MCP needs an Enterprise licence. Cloud has a 21-day trial with no payment details.

Side by side

Bonnard and Lightdash MCP servers, verified 16 Nov 2026
FeatureBonnardLightdash
Who answersThe AI client the person uses, with Bonnard’s contextThe AI client the person uses, with Lightdash’s semantic layer
Context the agent readsSchema, descriptions, joins, metrics, business terms, searchable valuesExplores, metrics, dimensions, descriptions, AI hints, verified content
Context through MCPexplore_schema, for any AI clientlist_explores, grep_fields, get_metadata and list_verified_content, for any AI client
Metric definitionsThe agent reads the SQL and writes the queryCompiled by Lightdash in run_metric_query. run_sql skips them, for users with the SQL Runner permission
ObserveRecords each tool call, from any AI clientMCP Analytics lists tool calls, for 90 days
DiagnoseFinds what is wrong in the activity recordIssues (beta), for Lightdash agent turns
ImproveThe data team’s answer goes into a draftPull requests on YAML models and agents
VerifyEvals: the draft against the live versionEvaluation suites for Lightdash agents, graded by an LLM
Run a querysql_query, one read-only statement on your databaserun_metric_query on explores, or run_sql
Charts and dashboardsrender_chart and render_dashboard, in the AI clientrender_chart, one chart from a metric query. Dashboards are saved in Lightdash
Who can connectUnlimited users, your customers’ users includedPeople with a Lightdash account, or an admin token with user attributes
Needed firstA data source: your database, or the sample dataA Lightdash project with YAML models. A warehouse alone gives SQL only
Databases and warehouses7, plus PostHog10, with Supabase through the Postgres connector
Tools for the agentUp to 1733 in the docs, 37 in the tool schema, by role and settings
How you payFor each call, no seatsFrom USD 3,000 a month for Cloud Pro, unlimited users, plus MCP usage
Free plan3,000 calls a monthNo MCP on Open Source without an Enterprise licence. A 21-day Cloud trial

When Lightdash is stronger

  • You want metric definitions enforced: run_metric_query compiles the SQL from your YAML, and users without the SQL Runner permission cannot query any other way. Bonnard gives the agent the metric’s SQL and does not rewrite the query.
  • You already keep metrics and dimensions in dbt or Lightdash YAML: Lightdash reads them from your repository, and the AI client gets them with no second copy.
  • You want context changes reviewed in git: models, the project context file, agents and evaluation suites are YAML, and run_ai_writeback lets an agent open a pull request against the dbt project (AI writeback).
  • You want agents to save and change charts and dashboards in Lightdash (create_content, edit_content), and to schedule them to Slack or email (create_scheduled_delivery).
  • You want agents to build a custom chart type (create_custom_chart_type) or a data app (generate_data_app). Data apps are an add-on with custom pricing (Lightdash pricing).

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

Lightdash. Settings > MCP > Connect gives the server URL and the setup for Claude Desktop, Claude Code, Codex, Cursor or VS Code (connect a client). In Claude, select Settings > Connectors > Add custom connector and enter this URL. In ChatGPT, create a connector with the URL and OAuth. The instance name is often app or eu1, and a self-hosted instance uses its own domain.

https://<your_instance_name>.lightdash.cloud/api/v1/mcp

Verified 16 Nov 2026 against Lightdash’s MCP docs, pricing, self-hosted MCP, connect a project, semantic layer, descriptions, user attributes, data access, visibility, Issues, evaluation suites, effective analytics, agent context, verified answers, agents as code, AI writeback, preview projects, tool schema and server instructions.

Get started

Start scaling accurate data today

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

Keep reading