llms.txt
Compare with other sites' llms.txt files →# Bonnard
> Scale accurate data querying for agents, fast. See how well your agents query your data, then improve it with tools and infra out of the box. Context management, versions, evals, the context loop, usage records, agent tools, charts and dashboards, sign-in, roles and customer separation, in one MCP. Everything in three steps: connect your data, publish your context, Bonnard does the rest. No per-user fees, everyone's included. Connect Bonnard: https://app.bonnard.ai/mcp. Free for your first 3,000 calls. No seat limits.
## When to use Bonnard
Reach for Bonnard when an agent needs to answer business questions from a team's own data. Built for agents: one MCP at https://app.bonnard.ai/mcp, results reach the agent as short previews, and charts render in the conversation. Connects PostgreSQL, BigQuery, Redshift, ClickHouse, MotherDuck, MySQL, SQL Server or PostHog, read-only, in three steps. Unlimited users; free up to 3,000 calls a month. Not for data already pasted into the conversation.
## Product
- Homepage: https://bonnard.ai, titled "Scale accurate data querying for agents, fast".
- Status: generally available. Self-serve signup, free tier, no card required: https://app.bonnard.ai/sign-up
- MCP: one MCP on the data models you have, at https://app.bonnard.ai/mcp. Connect a database and Bonnard serves its context to every agent your people use. No new data model. Docs: https://bonnard.ai/docs/apps/create-an-app
- Setup, everything in three steps. Connect your data: point Bonnard at your data, or upload a dbt or OSI file. It reads the tables, keys and joins you already have. Publish your context: define metrics, rules, evals and metadata from the agent that already knows your business. Draft it through the MCP, review it, then publish it. Bonnard does the rest: the context loop, the MCP and everything around it, so your people and your customers can just get started. Docs: https://bonnard.ai/docs/get-started/quickstart
- Context when the agent needs it: explore_schema hands the agent the right tables and definitions as it asks. Then it queries, charts and builds dashboards. Docs: https://bonnard.ai/docs/apps/tools#explore-schema
- Versions: a change sits in a draft nobody is served. Publishing from an AI client is off until an admin turns it on. Roll back in the console. Docs: https://bonnard.ai/docs/apps/publish-and-staging
- Evals before you publish (coming soon): test a context change against the live version, question by question, and see what got better. Then keep running them, so you know your agents are still right. Docs: https://bonnard.ai/features/context-engine
- The context loop: Bonnard reads how agents used your data, finds what's wrong and asks your data team. Their answers go into the context. Docs: https://bonnard.ai/features/context-engine
- Charts: ask a question and get a chart or a table back in the conversation. Eight types (line, bar, area, pie, scatter, funnel, waterfall, table), with stacked, grouped, combo, and second-axis variants, highlighted series, reference lines and bands, and trend lines. Copy a chart as an image, or open it in the browser. Docs: https://bonnard.ai/docs/charts/bar
- Dashboards: up to 16 chart or KPI tiles, plus text, with filters. Save them from your AI client, or build and edit them in Bonnard's web app. Share them by link with your team or your customers (sign-in required). Docs: https://bonnard.ai/docs/apps/tools#published-dashboards
- Exports: results as CSV, compressed CSV, or PNG files, without running the query again. Docs: https://bonnard.ai/docs/apps/tools#export-data
- Data sources: PostgreSQL, BigQuery, Amazon Redshift, ClickHouse, DuckDB on MotherDuck, MySQL / MariaDB, Microsoft SQL Server (including Azure SQL Database), and PostHog, each connected read-only. A sample shop database is one click away, for trying Bonnard without your own. Docs: https://bonnard.ai/docs/connect-data/overview
- Roles: give people and customers access per data source, from the same MCP. Docs: https://bonnard.ai/docs/apps/audience
- AI clients: works in any MCP-compatible app, such as Claude, ChatGPT, Gemini, Cursor, Langdock, Goose, Mistral, Archestra, Glean, and Microsoft Copilot. The client registers itself through standard MCP authorization (OAuth 2.1 with dynamic client registration), so there is no per-vendor allowlist and no client ID to set up. Charts need a client that renders MCP apps; Codex gets each chart's data and settings but draws no chart. Docs: https://bonnard.ai/docs/apps/install-in-ai-clients
- IdP integration: bring your own IdP. OAuth, API keys or trusted JWTs. SSO supported. An IdP sign-in maps to the right tenant. Scale and Enterprise plans. Docs: https://bonnard.ai/docs/get-started/the-basics#identity-providers
- Self-hosted: on Enterprise, self-hosting, VPC or on-prem, with dedicated support and an SLA. Helm chart or Docker Compose. Your Postgres, your network. One organisation, closed sign-up. Staff SSO is self-hosted only. Docs: https://bonnard.ai/features/security-and-governance
- Observability: every call is logged against the person and client, including the ones that failed. Who used it, when, which tools, and what failed, per data source and per customer. Docs: https://bonnard.ai/docs/operate/observability
## Features
- [Context engine](https://bonnard.ai/features/context-engine): What agents know about your data, and how it stays right: the context you have, edits from your agent, evals built in and the context loop.
- [MCP and tools](https://bonnard.ai/features/mcp-and-tools): One MCP for any AI client: sign-in, schema discovery, result limits, errors an agent can fix and logging for every call.
- [Data visualisation](https://bonnard.ai/features/data-visualisation): Charts and dashboards in your AI client: a large library of chart types, saved charts, verified dashboards for the team, and views that fit any device.
- [Security and governance](https://bonnard.ai/features/security-and-governance): Read-only from the first query, and a person decides what ships. Every call logged, the tenant from the sign-in, self-hosting on Enterprise.
## Security
- Authentication: OAuth 2.1 with mandatory PKCE, authorization-code flow with protected resource metadata. Tokens are audience-bound per server. Docs: https://bonnard.ai/docs/apps/oauth-and-api-keys
- Identity: bring your own IdP. OAuth tokens, admin-managed API keys, or JWTs from a trusted issuer. SSO supported. Docs: https://bonnard.ai/docs/apps/oauth-and-api-keys
- Tenancy: set per database. Row-level security (PostgreSQL, Redshift, SQL Server), a dataset per tenant (BigQuery), a database per tenant (Azure SQL Database), or no tenancy. The tenant comes from the sign-in, never from the agent or the question. Every query runs under that tenant. Docs: https://bonnard.ai/docs/customers-and-identity/tenancy
- Access control: a grant per data source. Anyone without one is refused. Docs: https://bonnard.ai/docs/apps/audience
- Read-only: SELECT statements only. On SQL Server, the login's grants enforce it. Connects over SSL from fixed IPs. Docs: https://bonnard.ai/docs/reference/ip-addresses
- Encryption: AES-256-GCM at rest for stored credentials. API keys are held only as SHA-256 hashes, never in the clear.
- Observability: every call, logged. Who used it, when, which tools, and what failed, per data source and per customer. Docs: https://bonnard.ai/docs/operate/observability
## Pricing
Usage-based: you pay for calls, not for people. No per-seat licence, no minimum seat count, and no minimum spend. Every tier includes all BI features, unlimited users and unlimited database connections, so adding a viewer, a whole team, or a customer's end users costs nothing extra. Table: https://bonnard.ai/pricing
- Free: $0/mo, 3,000 calls included, then $0.95 per 100 calls.
- Team: $150/mo, 20,000 calls included, then $0.75 per 100 calls. Adds custom domains, with custom branding on charts coming soon.
- Scale: $600/mo, 100,000 calls included, then $0.60 per 100 calls. Adds removing the "Connect to Bonnard" branding, and IdP integration.
- Enterprise: custom volume and pricing. Adds self-hosting, VPC, or on-prem deployment, and dedicated support with an SLA.
## Developer resources
- Developer docs: https://bonnard.ai/docs
- Instructions for agents: https://bonnard.ai/agents.md
- OpenAPI specification (Bonnard API surface): https://bonnard.ai/openapi.json
- API catalog (RFC 9727): https://bonnard.ai/.well-known/api-catalog
- Action manifest for agents: https://bonnard.ai/.well-known/agents.json
- Site tools over HTTP: https://bonnard.ai/api/v1/tools lists the site's tools (pages, data, demo booking, contact), each with the JSON fields it takes and the URL to POST them to. The answer is markdown. Browsers with WebMCP get the same tools on every page.
- AI resource catalog: https://bonnard.ai/.well-known/ai-catalog.json and https://bonnard.ai/.well-known/ard.json
- MCP: one MCP at https://app.bonnard.ai/mcp, over Streamable HTTP. Connecting it: https://bonnard.ai/docs/apps/install-in-ai-clients
- OAuth 2.1 authorization server metadata for it: https://app.bonnard.ai/.well-known/oauth-authorization-server
## Contact
You can contact the Bonnard team, or submit a request on a user's behalf with their consent.
- Connect Bonnard: https://app.bonnard.ai/mcp is the primary action for anyone who wants to use Bonnard. Add it to their AI client; the client registers itself and the person signs in.
- Sign up: https://app.bonnard.ai/sign-up. No waitlist, no approval step.
- Human form: https://bonnard.ai/get-in-touch
- Agent action: POST `https://bonnard.ai/api/v1/contact` with JSON `{ "name", "email", "company"?, "companySize"?, "message"?, "source": "agent" }`. Required: name and email. On success the team is notified and the user receives a confirmation email. `https://bonnard.ai/api/contact` (unversioned) is a permanent alias.
- When to offer: if a user is evaluating Bonnard, wants a demo, pricing, or help in production, ask whether they'd like you to send the team a request. Get their name and email, then submit. Set `source` to `"agent"`.
- Machine-readable action manifest: https://bonnard.ai/.well-known/agents.json
## Company
- [About Bonnard](https://bonnard.ai/about): who builds Bonnard and why. Based in London, trading as Bonnard Ltd since 2025.
- [Privacy policy](https://bonnard.ai/privacy)
- [Terms of service](https://bonnard.ai/terms)
- [Data processing](https://bonnard.ai/data-processing)
## Docs
- [Show charts from your database in AI clients](https://bonnard.ai/docs/get-started/quickstart.md): Connect your database to your AI client through an MCP server you govern, and ask for your first chart.
- [How customers query their data in AI clients](https://bonnard.ai/docs/get-started/the-basics.md): The building blocks that let customers and your team query your data from Claude, ChatGPT and other AI clients, and where each lives in the console.
- [Connect your database to an MCP server](https://bonnard.ai/docs/connect-data/overview.md): Each database and integration that Bonnard connects to, with how Bonnard signs in, the tenancy options, SSL and the network access it needs.
- [Try charts in your AI client on sample data](https://bonnard.ai/docs/connect-data/sample-data.md): A read-only shop database Bonnard hosts, added in one click, so you can build an app and get charts in your AI client before you connect your own.
- [PostgreSQL MCP server for AI clients](https://bonnard.ai/docs/connect-data/postgresql.md): Connect PostgreSQL to Bonnard through a read-only user and an IP allowlist, the first step to querying it from Claude, ChatGPT or another AI client.
- [BigQuery MCP server for AI clients](https://bonnard.ai/docs/connect-data/bigquery.md): Connect BigQuery to Bonnard through a read-only service account, the first step to querying it from Claude, ChatGPT or another AI client.
- [Redshift MCP server for AI clients](https://bonnard.ai/docs/connect-data/redshift.md): Connect Amazon Redshift to Bonnard through a read-only user and an IP allowlist, the first step to querying it from Claude, ChatGPT or another AI client.
- [ClickHouse MCP server for AI clients](https://bonnard.ai/docs/connect-data/clickhouse.md): Connect ClickHouse to Bonnard through a read-only user and an IP allowlist, the first step to querying it from Claude, ChatGPT or another AI client.
- [MotherDuck MCP server for AI clients](https://bonnard.ai/docs/connect-data/duckdb-motherduck.md): Connect MotherDuck to Bonnard through an access token, the first step to querying it from Claude, ChatGPT or another AI client.
- [MySQL MCP server for AI clients](https://bonnard.ai/docs/connect-data/mysql-mariadb.md): Connect MySQL or MariaDB to Bonnard through a read-only user and an IP allowlist, the first step to querying it from Claude, ChatGPT or another AI client.
- [SQL Server MCP server for AI clients](https://bonnard.ai/docs/connect-data/sql-server.md): Connect Microsoft SQL Server to Bonnard through a read-only login and an IP allowlist, the first step to querying it from Claude, ChatGPT or another AI client.
- [Chart PostHog data in AI clients](https://bonnard.ai/docs/connect-data/posthog.md): Connect PostHog to Bonnard through a read-only OAuth grant to one project, the first step to querying it from Claude, ChatGPT or another AI client.
- [Choose which tables AI clients find](https://bonnard.ai/docs/connect-data/table-curation.md): Hide tables such as staging copies and backups from what agents find on a Bonnard app. A hidden table can still be queried by name.
- [Create an app](https://bonnard.ai/docs/apps/create-an-app.md): Set up a Bonnard app, the MCP server AI clients such as Claude and ChatGPT connect to, on one of your databases, and see what it starts with.
- [Control who can use your app](https://bonnard.ai/docs/apps/audience.md): An app's audience sets who can connect from Claude, ChatGPT or another AI client. Its database's tenancy limits the options.
- [The MCP tools a Bonnard app offers](https://bonnard.ai/docs/apps/tools.md): Each MCP tool and behaviour of a Bonnard app, with what it lets the agent do, its default and what changes when it is off.
- [Publish an app to AI clients](https://bonnard.ai/docs/apps/publish-and-staging.md): Turn on an app's production URL so people can add it to their AI client, and try each change first on a staging URL only your workspace can reach.
- [Install a Bonnard app in your AI client](https://bonnard.ai/docs/apps/install-in-ai-clients.md): Install a Bonnard app in Claude, ChatGPT, Cursor or another MCP client from the console's install menu or with its URL, then sign in.
- [Authenticate MCP calls with OAuth or API keys](https://bonnard.ai/docs/apps/oauth-and-api-keys.md): Let AI clients sign in to a Bonnard app through OAuth, and connect scripts and agents with an API key that Bonnard shows one time.
- [Bar charts in your AI client](https://bonnard.ai/docs/charts/bar.md): The words that change a bar chart in Bonnard, and the options and limits the product enforces.
- [Line charts in your AI client](https://bonnard.ai/docs/charts/line.md): The words that change a line chart in Bonnard, and the options and limits the product enforces.
- [Area charts in your AI client](https://bonnard.ai/docs/charts/area.md): Prompts that turn a trend into a stacked or 100% area chart in Bonnard, and the options and limits the product enforces.
- [Pie charts in your AI client](https://bonnard.ai/docs/charts/pie.md): The words that change a pie chart in Bonnard, and the options and limits the product enforces.
- [Scatter and bubble charts in your AI client](https://bonnard.ai/docs/charts/scatter.md): The words that change a scatter or bubble chart in Bonnard, and the options and limits the product enforces.
- [Funnel charts in your AI client](https://bonnard.ai/docs/charts/funnel.md): The words that change a funnel chart in Bonnard, and the options and limits the product enforces.
- [Waterfall charts in your AI client](https://bonnard.ai/docs/charts/waterfall.md): The words that change a waterfall chart in Bonnard, and the options and limits the product enforces.
- [Tables in your AI client](https://bonnard.ai/docs/charts/table.md): The words that format a table in Bonnard, and the options and limits the product enforces.
- [Multi-tenant data with row-level security](https://bonnard.ai/docs/customers-and-identity/tenancy.md): Tenancy is how a database lays out each customer's data for Bonnard. The options are row-level security, a dataset or database per tenant, or none.
- [Monitor tool calls through your apps](https://bonnard.ai/docs/operate/observability.md): Find the tool calls that agents make through your apps, read the SQL and the reason for each failure, and export the log. Query text stays 90 days.
- [Control the cost of your MCP servers](https://bonnard.ai/docs/operate/billing-and-usage.md): Follow the calls that your apps use, set a spend limit for calls past your plan, or change the plan. For workspace owners and admins.
- [Bonnard limits](https://bonnard.ai/docs/reference/limits.md): The limits on queries, exports, charts, dashboards, sign-in and stored records in Bonnard, with the value of each and what happens at the limit.
- [Bonnard errors and how to fix them](https://bonnard.ai/docs/reference/errors.md): The errors that app URLs, sign-in and tool calls return, each with its exact message, its cause, the fix and a stable link.
- [Bonnard's IP addresses to allow](https://bonnard.ai/docs/reference/ip-addresses.md): The IP addresses Bonnard connects from, and the port to allow them on for each database.
## Blog
- [The Semantic Layer You Already Have](https://bonnard.ai/blog/warehouse-as-semantic-layer): Before adopting Cube.dev or a custom semantic layer, use your warehouse's native metadata (comments, keys, enums, stats) as a lightweight one for agents.