Skip to content

Exposing Analysis Over MCP

Charts and networks are great for a quick look, but some questions deserve a longer, unstructured conversation — "based on my blood panel results and logged supplement intake, do you recommend making any modifications to my supplement stack?" is the kind of thing no single dashboard is built to answer on its own. Every analysis method behind the Analysis and Signal pages is also exposed as an MCP tool, so your AI assistant can run the real numbers instead of guessing from vibes.

See Setting up MCP to connect an agent first.

Finding your way around

  • catalog — a one-shot inventory of every metric you track (name, category, type, unit, recent activity). The starting point for any analysis conversation.
  • list_categories / describe_metric — narrower lookups when your assistant already knows roughly what it's after.

Pulling raw data

  • query_measurements — raw measurements for a list of metrics over a date range.
  • recent — the N most recent measurements per metric.
  • fetch_recent_measurements — a bulk dump of everything logged in the last N days, across every metric if you don't narrow it down.

Running the analysis

  • summarize — summary statistics for a set of metrics over a date range.
  • correlate — a single-pair correlation, optionally lagged by a number of days.
  • correlate_matrix — every pairwise correlation across a list of metrics in one call, with the strongest associations ranked at the top.
  • compare_periods — how a metric changed between two date ranges.
  • anomalies — flags measurements that don't fit the recent pattern, with an adjustable sensitivity.

Because these are the same tools the Analysis and Signal pages call under the hood, an answer your assistant gives you is backed by the same statistics you'd see if you built the chart yourself — not an approximation.

A good way to use this

Rather than asking a single narrow question, try handing your assistant an open-ended one and letting it explore: "Look at everything I've logged and tell me what's most strongly connected to my sleep score." It can pull the catalog, run correlate_matrix across your metrics, and walk you through what it finds — the same kind of session the Signal page's network view is built to support, just in conversation instead of a graph.