Databuddy MCP server

What it does

Databuddy is an open-source, cookie-free analytics product, and this is the MCP endpoint its makers run at api.databuddy.cc. Any client that speaks Streamable HTTP can connect it to a Databuddy workspace. From there an assistant can pull traffic and error figures through typed queries, check the analytics schema when a field name is unclear, and read conversion and drop-off for goals and funnels that already exist in the dashboard.

The newer parts of the product are reachable too. You get the investigations and published findings from Databunny, the built-in analyst, and can post a follow-up question on an investigation. Feature flags and their targeting rules can be listed and edited, as can annotations on the timeline and short links. Most of these objects can also be created, and goals, annotations and links deleted. There is also a feedback tool that files a bug report with the Databuddy team.

Edits work in two steps. A create, update or delete call returns a preview, and nothing is written until the same call comes back with confirmed set to true. Investigation replies and feedback skip that step. The tool list a client sees follows the granted scopes, so a connection holding only read:data is a read-only analytics helper. The docs table counts 32 tools, and the server card at api.databuddy.cc also names a resource, databuddy://guide, which holds a markdown guide to query conventions.

What it needs

Claude and Claude Code sign in with OAuth and need no key. Cursor, Windsurf and other clients send an API key in the x-api-key header (or as a Bearer token). The quickest way to get one is Organization Settings, Integrations, Databuddy MCP, which creates a dedicated key with the scopes you pick and shows the secret once.

Limits

• Size cap Each get_data query returns at most 20 rows, and a batch holds 2 to 10 queries.
• Size cap Time series span at most 400 days, 30 days with hour buckets and 1 day with minute buckets. Dates default to the last 30 days.
• Rate limit Per-tool limits per credential: 60 calls a minute for list tools, 30 for get_data, 10 for creates and deletes of goals and funnels, 5 an hour for submit_feedback.
• Rate limit OAuth connections allow at most 20 requests in flight per account and app; more get a 429 with Retry-After.
• Not covered Signup and conversion answers need goals or funnels already set up in the dashboard. It cannot define tracking or read other analytics tools.
• Cost It cannot start a new Databunny analysis. That is done in the dashboard, which shows a $1 price before submitting.
• Not covered Connections limited to specific websites cannot reach organization-wide feature flags.
• Limit One JSON-RPC message per POST: batch arrays get a 400 and bodies over 1 MB a 413. There is no SSE or WebSocket transport.

Each get_data query returns at most 20 rows and a batch holds 2 to 10 queries; time series span at most 400 days (30 for hourly, 1 for minute buckets). Per-tool rate limits run from 60 calls a minute for list tools down to 10 for creates and deletes. Answers about signups depend on goals or funnels already set up in the dashboard; the server cannot start a fresh Databunny analysis, which is done in the dashboard for a $1 charge. Website-limited connections cannot reach organization-wide flags. No SSE or WebSocket transport. The self-hosting guide calls self-hosting still evolving.

Who it suits

Fits teams that already track their sites with Databuddy and want to ask traffic, error and conversion questions in Claude, Cursor or similar without opening the dashboard. Because it is the vendor's own hosted endpoint, data stays on Databuddy's side and setup is a URL plus sign-in or a key. Start with a read:data key or the default OAuth permissions for pure reading; grant workspace, flag or link scopes only to clients you trust to change those. Of no use without a Databuddy workspace: it reads nothing from other analytics tools.