Connect Claude and other AI assistants
Let Claude, Cursor or any MCP client read your analytics: the URL, the steps and configs to copy, and what to ask.
Little Friend runs an MCP server, so Claude, Cursor and other AI assistants can read your reports. Every connection is read only, reaches only the projects you allow, and can be revoked at any time. It never sees raw IP addresses, user agents or anything typed on your site.
Claude on the web and Claude Desktop
No key needed. Claude signs in to Little Friend, and you choose what it can read.
- Add a custom connector. In Claude, open Settings, then Connectors, and choose Add custom connector.
- Paste the MCP URL. Name it Little Friend. Leave the OAuth client ID and secret empty.
MCP URL https://api.littlefriend.io/mcp - Connect and pick projects. Choose Connect. Little Friend opens: sign in, pick the projects Claude may read, and approve.
On Team and Enterprise plans, an owner may need to add the connector for the organization first.
Claude Code
Run this in a terminal. It needs no key either.
claude mcp add --transport http littlefriend https://api.littlefriend.io/mcpThen run /mcp in Claude Code, pick littlefriend and sign in. Add --scope user to use it in every project.
Cursor and other clients, with a read key
For clients that can't sign in, create a project read key in the dashboard under Settings, AI assistants, Create a read key. It reads one project and is shown once, so copy it right away. Put it where the configs below say lfr_YOUR_READ_KEY.
{
"mcpServers": {
"littlefriend": {
"url": "https://api.littlefriend.io/mcp",
"headers": {
"Authorization": "Bearer lfr_YOUR_READ_KEY"
}
}
}
}claude mcp add --transport http littlefriend https://api.littlefriend.io/mcp \
--header "Authorization: Bearer lfr_YOUR_READ_KEY"Avoid --scope project with a key: it writes the key into .mcp.json, which is usually checked in.
A client that only runs local servers can reach the MCP URL through the mcp-remote bridge, for example from Claude Desktop's config file. It runs with npx, so Node.js must be installed.
{
"mcpServers": {
"littlefriend": {
"command": "npx",
"args": [
"-y",
"mcp-remote",
"https://api.littlefriend.io/mcp",
"--header",
"Authorization:${AUTH_HEADER}"
],
"env": {
"AUTH_HEADER": "Bearer lfr_YOUR_READ_KEY"
}
}
}
}What an assistant can do
- Reports and breakdowns. Headline numbers, sources, pages, countries, devices and campaigns, next to the period before.
get_reportget_breakdown - What changed. The biggest movers since the week before, or since a year earlier, with both values.
what_changed - Goals, funnels, paths and flows. Conversions with their denominators. Saved and one-off funnels, paths, flows and session timelines in journey mode.
list_goalslist_funnelsget_funnelget_pathsget_flow_startsget_flowsget_session - Crawlers and AI agents. Which bots and AI agents read which pages, with evidence, and the people AI assistants send.
get_agent_activity - Data health. Whether the install works and the numbers are fresh, down to the newest events and rejects.
data_healthget_live - Definitions and exports. The definition behind a metric, and any report as CSV.
explain_metricexport_report
When a number is unavailable, the assistant is told so, instead of getting a zero. Journey reports need a project in journey mode, and the assistant is told which tools work in yours.
Questions to ask
- What changed on the pricing page last week?
- Which AI crawlers read our docs last week?
- How many people did ChatGPT and Perplexity send us this month?
- Which referrers grew the most compared with the month before?
- What is our signup conversion rate, and how is it counted?
- Is the data fresh, and is anything being rejected?
Two ready-made prompts come with it: weekly_summary for a short summary of the last complete week, and ai_visibility_review for which AI crawlers and agents read your site.
Cutting one off
Revoke a signed-in connection under Settings, AI assistants, Connected apps, or remove the connector in Claude. Revoke a read key under Settings, Keys. Either stops working at once. Questions? Write to hello@littlefriend.io.