# Bonnard vs Zenlytic: MCP server features compared

> Bonnard and Zenlytic MCP servers compared: what the agent knows about your data, how wrong answers get fixed, who can connect and pricing.

With Zenlytic (zenlytic.com), Claude or ChatGPT passes the question to Zenlytic's agent, Zoë, and a checker validates Zoë's SQL. With Bonnard (bonnard.ai), Claude or ChatGPT reads the context and writes the query.

## 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](/features/context-engine)). 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.

Zenlytic keeps its context for Zoë. The context is a semantic model of views, measures, dimensions and relationships, plus a system prompt and skills ([context surfaces](https://docs.zenlytic.com/getting-started/context-surfaces)). A field can carry a description, synonyms and a `zoe_description` that only Zoë reads. A dimension marked `searchable: true` has its values indexed. On each question, Zoë's `search fields` tool finds the measures, dimensions and joins that apply, with their SQL definitions ([how Zoë works](https://docs.zenlytic.com/getting-started/how-zoe-works)).

Zoë writes the SQL, and the [Clarity Engine](https://docs.zenlytic.com/using-zenlytic/clarity_engine) checks it against the model before it runs. Where the SQL uses a modeled measure or dimension, the engine applies the governed definition. Row and column rules apply to each query. Logic that is not in the model becomes a Dynamic Field for that question and shows as unverified. Modeled fields show a green check. The definitions are applied where Zoë uses them, and Zoë can still calculate a number outside the model.

Zenlytic's [MCP docs](https://docs.zenlytic.com/mcp/connecting-to-zenlytic) name one tool, `ask_zoe`. It sends the question to Zoë and returns the answer, a link to the conversation in Zenlytic and any data or charts. Claude or ChatGPT does not read the model, the system prompt or the skills. It gets Zoë's answer.

### How wrong answers get found and fixed

Bonnard runs a loop of four steps on how agents use your data. Zenlytic records the conversations and lets Zoë make the fix, and a person does the diagnosis and the testing.

[Image: The context loop in four steps: observe, diagnose, improve, verify.]

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

Each MCP conversation shows in Zenlytic's conversation history, tagged MCP, and workspace admins can see the question and the answer ([MCP docs](https://docs.zenlytic.com/mcp/connecting-to-zenlytic)). Admins [review chats and user feedback](https://docs.zenlytic.com/interactive-demos/workspace-administration-for-admins/review-chats-and-feedback) in one place.

**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.

[Image: Bonnard's activity record over 30 days. It finds three definitions of net revenue across eight people, a search for gift cards that found nothing, sessions that took nine calls, an undefined column error and a table that stopped in May.]

On Zenlytic, a person does this step. Zenlytic's guide to [fixing Zoë's mistakes](https://docs.zenlytic.com/getting-started/fixing-zoes-mistakes) has five steps, and the first three are to read the fields Zoë searched and the SQL that ran, find the failure and classify it. Zoë audits the model on request and returns findings without saving. Zenlytic documents no suggestions generated from agent activity.

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

The guide's fourth step is to add the smallest piece of context that prevents the failure. Zoë can make that fix. With Context Editing on, a user with the Develop or Admin role asks Zoë in chat to change the model. Zoë drafts the edit, validates the model and commits it to the current branch ([Context Editing](https://docs.zenlytic.com/data-modeling/asking-zoe-for-recommendations)). [Context Manager](https://docs.zenlytic.com/data-modeling/context_manager) compares a branch with production, shows validation results and deploys the branch.

**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.

[Image: Three questions, each run five times against the live context and against the draft. Two get better and one shows no clear difference.]

Zenlytic documents no evals. Zoë runs a sample query on each new or changed field, and the guide's last step is to ask the question again. Ask Zenlytic how to test a set of questions against a branch before you deploy it.

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.

Zenlytic's loop depends on a person who reads conversations and asks Zoë for a fix, and it improves Zoë's answers. Bonnard's loop starts from the activity record, and improves what Claude, ChatGPT or Cursor read before they write a query.

### 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.

Zenlytic starts from a workspace with a connection to your warehouse. Each table that you import becomes a view, with dimensions made from its columns ([start here](https://docs.zenlytic.com/getting-started/start_here)). Zenlytic states that you do not need a finished data model first. The model lives in a git repository, which Zenlytic manages by default. An MCP conversation uses the branch that the workspace is set to when the conversation starts, development branches included ([MCP docs](https://docs.zenlytic.com/mcp/connecting-to-zenlytic)). Check the branch before you act on an answer.

### 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](/docs/access/tenancy)).

On Zenlytic, each person needs a Zenlytic account with access to the workspace and the chat permission ([MCP docs](https://docs.zenlytic.com/mcp/connecting-to-zenlytic)). Each role has chat except Restricted ([user roles](https://docs.zenlytic.com/administration/user_roles)). The person signs in through OAuth and selects a workspace, or sends a personal access token in an `Authorization` header. Creating a token needs the Admin or Organization Admin role. Zenlytic serves your customers' users through [signed embedding](https://docs.zenlytic.com/embedding/signed_embedding) of its dashboards and of Zoë's chat in your application. Ask Zenlytic if an embed user can connect over MCP.

### 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](/pricing)).

Zenlytic publishes its pricing model and no amounts. A subscription combines platform, seat-based and usage-based pricing ([Zenlytic vs ThoughtSpot](https://zenlytic.com/zenlytic-vs-thoughtspot)) and needs a custom quote ([Zenlytic vs WisdomAI](https://zenlytic.com/zenlytic-vs-wisdom)). The address `zenlytic.com/pricing` shows [Book a Demo](https://zenlytic.com/book-a-demo), where you book a 30-minute call. Zenlytic offers a free trial, in which you connect your warehouse and try Zoë on your own data. A Zenlytic [blog post](https://zenlytic.com/blog/data-governance-tools) names a 14-day trial. Ask Zenlytic if the trial and your plan include the MCP server.

## Side by side

Table: Bonnard and Zenlytic MCP servers, verified 16 Nov 2026

| Feature | Bonnard | Zenlytic |
| --- | --- | --- |
| Who answers | The AI client the person uses, with Bonnard's context | Zenlytic's agent, Zoë, through `ask_zoe` |
| Context the agent reads | Schema, descriptions, joins, metrics, business terms, searchable values | Semantic model, descriptions, synonyms, searchable values, system prompt, skills |
| Context through MCP | `explore_schema`, for any AI client | No tool reads it. `ask_zoe` returns Zoë's answer |
| Metric definitions | The agent reads the SQL and writes the query | The Clarity Engine checks each query against the model and applies the definition where Zoë uses a modeled field. Other logic shows as unverified |
| Observe | Records each tool call, from any AI client | Conversation history, tagged MCP, and user feedback |
| Diagnose | Finds what is wrong in the activity record | By hand, from Zenlytic's guide |
| Improve | The data team's answer goes into a draft | Zoë edits the model on a branch when a person asks |
| Verify | Evals: the draft against the live version | No evals documented. Zoë runs a sample query on each field it changes |
| Run a query | `sql_query`, one read-only statement on your database | `ask_zoe`. Zoë writes the SQL and the Clarity Engine checks it |
| Charts and dashboards | `render_chart` and `render_dashboard`, in the AI client | `ask_zoe` returns data and charts, in a result view in AI clients with MCP Apps |
| Who can connect | Unlimited users, your customers' users included | People with a Zenlytic account and the chat permission |
| Needed first | A data source: your database, or the sample data | A Zenlytic workspace with a warehouse connection and imported tables |
| Databases and warehouses | 7, plus PostHog | [12](https://docs.zenlytic.com/data-sources/integrations) |
| Tools for the agent | Up to 17 | One named in the MCP docs, `ask_zoe` |
| How you pay | For each call, no seats | Custom quote: platform, seat-based and usage-based pricing, no amounts published |
| Free plan | 3,000 calls a month | No free plan published. A free trial on your own data |

## When Zenlytic is stronger

- You want each query checked against the model: the [Clarity Engine](https://docs.zenlytic.com/using-zenlytic/clarity_engine) validates Zoë's SQL, applies the governed definition where a modeled field is used, and marks the rest as unverified. Bonnard gives the agent the metric's SQL and does not rewrite the query.
- You already have a semantic layer in dbt: Zenlytic reads a [dbt MetricFlow](https://docs.zenlytic.com/data-modeling/dbt_metricflow) project from a GitHub repository and maps its semantic models, dimensions and measures.
- You keep context in git: the model and skills are files in a repository, with development branches and a deploy step in [Context Manager](https://docs.zenlytic.com/data-modeling/context_manager). A commit that Zoë makes records the person who asked for it as co-author ([changelog](https://docs.zenlytic.com/changelog)).
- You want the agent to edit its own context from chat: Zoë saves a synonym, a description or a new measure to a branch, after validation ([Context Editing](https://docs.zenlytic.com/data-modeling/asking-zoe-for-recommendations)).
- You want scheduled reports and files: Zoë builds spreadsheets, presentations, PDFs and HTML dashboards as [artifacts](https://docs.zenlytic.com/using-zenlytic/artifacts), and Zenlytic sends them to email or Slack on a schedule.

## Add each server to your AI client

**Bonnard.** Create a workspace, add Bonnard to your AI client, then add a data source ([quickstart](/docs/get-started/quickstart)). Use the install links for Claude, Cursor and ChatGPT in the console at `app.bonnard.ai`, or add this URL.

```text
https://app.bonnard.ai/mcp
```

**Zenlytic.** In Claude.ai, select **Settings**, then **Connectors**, then **Add** and **Add custom connector**. Paste the URL for your Zenlytic instance and select **Add**. Then select **Connect**, sign in to Zenlytic, select a workspace and approve the connection ([Claude.ai guide](https://docs.zenlytic.com/mcp/connecting-to-zenlytic/claude-ai)). Accounts at `app.zenlytic.com` use this URL, and accounts at `eu.zenlytic.com` use `https://mcp-eu.zenlytic.com/mcp`. A company-specific URL on the US instance takes the form `https://<company>.mcp.zenlytic.com/mcp`.

```text
https://mcp.zenlytic.com/mcp
```

In Claude Code, run this command, then run `/mcp`, select **zenlytic** and select **Authenticate** ([Claude Code guide](https://docs.zenlytic.com/mcp/connecting-to-zenlytic/claude-code)). Then sign in through the browser and select a workspace.

```bash
claude mcp add --transport http zenlytic https://mcp.zenlytic.com/mcp
```

Verified 16 Nov 2026 against Zenlytic's [MCP docs](https://docs.zenlytic.com/mcp/connecting-to-zenlytic), [MCP overview](https://docs.zenlytic.com/mcp/overview), [Claude.ai guide](https://docs.zenlytic.com/mcp/connecting-to-zenlytic/claude-ai), [Claude Code guide](https://docs.zenlytic.com/mcp/connecting-to-zenlytic/claude-code), [how Zoë works](https://docs.zenlytic.com/getting-started/how-zoe-works), [context surfaces](https://docs.zenlytic.com/getting-started/context-surfaces), [Clarity Engine](https://docs.zenlytic.com/using-zenlytic/clarity_engine), [fixing Zoë's mistakes](https://docs.zenlytic.com/getting-started/fixing-zoes-mistakes), [review chats and feedback](https://docs.zenlytic.com/interactive-demos/workspace-administration-for-admins/review-chats-and-feedback), [Context Editing](https://docs.zenlytic.com/data-modeling/asking-zoe-for-recommendations), [Context Manager](https://docs.zenlytic.com/data-modeling/context_manager), [start here](https://docs.zenlytic.com/getting-started/start_here), [user roles](https://docs.zenlytic.com/administration/user_roles), [embed roles](https://docs.zenlytic.com/embedding/permissions_in_embedding), [signed embedding](https://docs.zenlytic.com/embedding/signed_embedding), [data sources](https://docs.zenlytic.com/data-sources/integrations), [dbt MetricFlow](https://docs.zenlytic.com/data-modeling/dbt_metricflow), [artifacts](https://docs.zenlytic.com/using-zenlytic/artifacts), [changelog](https://docs.zenlytic.com/changelog), [Book a Demo](https://zenlytic.com/book-a-demo), [trial blog post](https://zenlytic.com/blog/data-governance-tools), [comparison with ThoughtSpot](https://zenlytic.com/zenlytic-vs-thoughtspot) and [comparison with WisdomAI](https://zenlytic.com/zenlytic-vs-wisdom).