Does it work in Cursor's free plan?

MCP is a Cursor feature, not a DataBull one. If your Cursor can use MCP servers, it can use DataBull: it's a standard Streamable HTTP server with a header.

Global or project config?

Global (~/.cursor/mcp.json) if you want your numbers in every project. Project (.cursor/mcp.json) if only one repo should see them; keep that file out of git.

What about VS Code, Codex and Windsurf?

npx databull@latest mcp prints the config for each. VS Code uses servers with "type": "http", Windsurf uses serverUrl, Codex uses [mcp_servers.databull] in ~/.codex/config.toml. See the docs.

Analytics MCP server for Cursor

Short answer

Add DataBull to ~/.cursor/mcp.json (or .cursor/mcp.json in one project) as a remote server: the URL is your DataBull address plus /mcp, with the header Authorization: Bearer dbk_…. npx databull@latest mcp --client cursor creates the read-only key and prints the block, filled in.

Get the config, filled in

npx databull@latest mcp --client cursor

It creates a read-only key, checks it works and prints the block below with your URL and key. On DataBull Hosted, or without the CLI, create a key under Settings → API keys & AI agents: the dashboard shows the same block.

The config

Put it in ~/.cursor/mcp.json to have DataBull in every project, or in .cursor/mcp.json in one project only:

{
  "mcpServers": {
    "databull": {
      "url": "https://YOUR-DATABULL-URL/mcp",
      "headers": { "Authorization": "Bearer dbk_YOUR_KEY" }
    }
  }
}

If the file already has other servers, add databull inside the existing mcpServers object. A project file holds your key, so add .cursor/mcp.json to .gitignore.

Turn it on

Open Cursor's MCP settings (under Customize in the sidebar in current versions) and check that databull is listed and switched on. By default Cursor asks before it runs a tool; every DataBull tool only reads, so allowing them is low-risk.

What the agent can read

DataBull gives Cursor 14 read-only tools. The ones you'll use most while building:

Tool	Good for
`get_breakdown`	Pages, landing pages, referrers or campaigns ranked by revenue or conversion.
`get_engagement`	Reading time and scroll depth per page, buyers against everyone else.
`get_funnels`	Your saved funnels, step by step, with drop-off and the revenue of visitors who reached each step.
`compare_periods`	Before and after a change, with the sources that moved most.
`list_sales` / `get_visitor`	Recent sales with their source, and one buyer's full path to paying.

The full list is on the MCP server page. Every tool takes a site by its domain and a period, and the answers use the same attribution model as your dashboard.

Prompts to try

"Which referrer made me the most money this month?"
"Which landing page converts best, and how far do buyers scroll on it?"
"Show my last 5 sales and the page each buyer landed on."
"Compare this week with last week for example.com."
"How many visits does it take people to buy?"

Watch revenue while you ship

Cursor is where you change the product, so ask before and after you change it:

Before you edit a page
"How does /pricing convert over the last 30 days, and what do buyers read before they pay?" gives you a baseline from get_breakdown and get_engagement.
On launch day
"Who's on the site right now, and where from?" (get_realtime), and "which source sent today's sales?" (list_sales).
A week later
"Compare the 7 days since the change with the 7 days before" (compare_periods). You see whether the new page made more money, not just more visits.

Leave a note on the dashboard chart when you ship, so the change shows up next to the numbers.

If it doesn't connect

Server shows an error or red dot: check the URL ends in /mcp and starts with https.
401 errors: the header value must be Bearer dbk_… with the space. "Unknown API key" means the key belongs to another DataBull; "revoked" means make a new one.
No tools appear: check the JSON is valid (a missing comma is the usual cause), then restart Cursor.

Ask Cursor which page converts