Lighthouse MCP and API
Lighthouse MCP: your metrics in Claude and Cursor
Lighthouse became something you can build in, not just check. Create and update metrics, pull their history or act on an alert, right from Claude Code or Cursor, with the same actions available as a plain REST API. Plus a long list of improvements to the metric builder.
Cursor · connected to Lighthouse via MCP
You: Create a metric for conversion in the last 3 hours vs the 30-day same-weekday average, alert if it's down more than 2 standard deviations
vs 30-day same-weekday avg · alerts beyond 2 STD
Lighthouse MCP
Build and act on metrics from Claude Code or Cursor
Turn it on in Settings and create a metric, update one that already exists, pull its history, or acknowledge an alert, without leaving your editor. An org admin can turn it on and create a key in a few minutes. Read the setup docs.
Terminal · calling the Lighthouse public API
curl .../v1/metrics/153/matches \
-H "x-api-key: lh_..."
{ "matched": true, "since": "2h ago" }
Public API
The same actions, as a REST API
Everything MCP can do is also a plain REST endpoint: list your metrics, run one, check its history, or acknowledge an alert with your API key. Wire it into a script, a dashboard or an internal tool, no AI client required. Read the API docs.
Also shipped
Sharper metrics, wider data sets
- Any percentile, not just the average. P90 and P95 across every data source, plus any custom percentile from 1 to 99, for teams that care about the tail of a distribution.
- Segmented thresholds for SQL metrics. The threshold follows your segments, so a problem in one segment doesn't get averaged away by the rest.
- NOT on a filter group. Exclude a combination instead of only including one, like NOT (Platform = Facebook and OS = iOS).
- Formulas inside formulas. A calculated column can reference another calculated column, so ratios and composite metrics need no workaround.
- Trend comparison on calculated columns. Show how installs compared to their own 30-day average in the same alert that fired on revenue.
- A SQL assistant that knows your warehouse. While you build a SQL metric it looks up your actual tables and columns and tries the query.
- Log scale on the trend chart. Metrics with a wide range still read clearly in the test view.
- Wide tables, handled. The column cap on a data set went from 100 to 300, and you can turn many columns on or off at once.
- See before you commit. A live preview on the metric form shows the table, type and columns before you save.
- Saved filters on the metrics list. Set up a view once and it's there next time.
- Lighthouse on your phone. Tables, dialogs and metric views now work on mobile.
