Your agent is now a data analyst.
Every Millimetric project ships with a native Model Context Protocol server. Claude, OpenClaw, Hermes, Cursor, Cline — anything that speaks MCP can both write events as users move and answer questions in natural language. One key, two-way.
{
"mcpServers": {
"millimetric": {
"url": "https://api.millimetric.ai/mcp",
"headers": {
"Authorization": "Bearer rk_live_…"
}
}
}
}Drop this into…
- Claude Desktop
~/Library/Application Support/Claude/claude_desktop_config.json - Cursor
.cursor/mcp.json - OpenClaw
openclaw.toml - Hermes
hermes.json - Cline
MCP marketplace UI - Continue
.continue/config.json
Any other MCP-compliant client works too — that's the whole point of the protocol.
MCP tools shipped
Tools an LLM actually likes.
Strict JSON schemas. Clear descriptions. No prompt-engineering tax.
- track_eventWrite a single event with arbitrary props.
- track_batchWrite up to 1,000 events in one call.
- get_statsPageviews, uniques, deltas for a metric + period.
- top_sourcesRanked source/medium breakdown for a window.
- top_pagesRanked page list for a window.
- query_eventsFiltered, paginated raw events (SQL-ish predicates).
- list_projectsProjects this key can read or write.
On ClawHub
One command, agent ready.
We publish three OpenClaw skills to ClawHub that turn the API and MCP server into native agent capabilities.
Millimetric Track
millimetric-trackEmit events from any agent — track, identify, batch, GDPR forget. Wraps /v1/track and friends.
clawhub install millimetric-track
needs sk_live_…
Millimetric Query
millimetric-queryRead analytics from the CLI — top sources, stats, raw events, the FB social-vs-paid split.
clawhub install millimetric-query
needs rk_live_…
Millimetric MCP Setup
millimetric-mcp-setupOne-shot wiring for Claude, Cursor, Inspector, OpenClaw — single-project and account-wide.
clawhub install millimetric-mcp-setup
needs rk_/sk_/ak_
First time on ClawHub?
Install the CLI, log in, then grab all three at once.
npm i -g clawhub clawhub login clawhub install millimetric-track millimetric-query millimetric-mcp-setup
Server-only keys
MCP accepts only rk_live_* (read) and sk_live_* (read + write) — keys you generate server-side and paste into your agent's config. pk_live_* keys, which ship in browser JS, are rejected at /mcp by design.
No PII surface
Default events contain no cookies, no IPs at rest, no fingerprints. Agents can't accidentally exfiltrate what we don't store.
Audit log on every call
Every MCP call is logged with the key prefix, IP and timestamp. Visible in the dashboard and exportable via API.
What this looks like in practice.
How did we do this week vs last week, in plain English?
I'll check.
→ tool: get_stats { metric: "signups", period: "7d", compare: "prev" }
Signups: 482 this week, up 12.4% from last. Top source was Hacker News (181). Mobile share jumped from 21% to 34% — worth a look.
Build the agent. We'll keep the books.
Free tier covers 100k events/mo and full MCP access. No dashboard required, ever.