Skip to main content
Use it

Read the Claude activity views

What the Adoption, People, Tools and connectors, and estimated cost views show, and how the period and Counts through stamp work.

Written By The Lemhi Product Team

Last updated 10 days ago

For anyone at the MSP reading a client's Claude activity. This covers what each view counts, what the labels mean, and how current the numbers are.

The views are on the client's AI Usage page, Claude tab, under the status header. They show while the status is Live, Stopped reporting or Rejected activity. For any other status, see Claude connection statuses and fixes.

Period and freshness

The period control above the status header switches between Last 7 days, Last 30 days (the default) and Last 90 days. A period is whole days in UTC, ending today. The summary tiles, Adoption, People, and Tools and connectors follow it. Models used and the estimated cost don't: they have fixed windows, described below.

Lemhi rolls up Claude's activity hourly.

  • The page header shows "Claude counts through [date and time]". If that time is more than two hours old, the stamp turns stale: the rollup is behind.

  • Each view shows "Counts through [date and time]" for its own numbers.

  • Before the first rollup, the page header reads "Claude counts update hourly", each view reads "Counts update hourly", and the views show no activity yet.

"Counts through" is not the same as the last activity received, which the status header shows.

Summary tiles

Above the tabs, four tiles replace the Microsoft 365 tiles: People active, Prompts, Tools and connectors in use (for the period), and Estimated API cost (last 30 days).

Adoption

  • Active in the last [N] days: people who sent at least one prompt in the period.

  • Change from the previous period: the signed change in active people against the same-length period just before.

  • Prompts: total prompts in the period.

  • Daily active people: people with prompts on each day.

  • Models used: share of prompts by model over the last 30 days, shown as count and percent on each bar. Model ids read as names, such as Claude Opus 4.6. A model Lemhi doesn't recognize shows as Claude sent it.

People

Tenant Admins only. Other users see "Ask an MSP admin to see per-person activity."

One row per person, sorted by prompts: Person, Prompts, Days active, Last active, Models, Daily prompts, and First seen, with a New tag for anyone first seen in the last 30 days. A person shows by name and department when exactly one of the client's employees has that email. Otherwise the email shows.

Search by name or email. Select a row to open the person's daily activity, prompts by model, and the tools and connectors they used. No cost is shown for any person, here or anywhere else.

Tools and connectors

Rows list in three sections:

Section

What's in it

Connectors

The client's connectors, by server name. Linear shows as Linear whichever server name Claude reports

Claude internals

Claude's own connectors: Claude memory, Remote devices, Scheduled tasks

Built-in tools

Claude's own tools, each under a plain name, such as Run terminal commands, Web search, Read files, Open web pages, and Built-in browser. A tool without a plain name shows its raw name

Columns are Tool or connector, Uses, People, First seen, and Failed or denied. The last shows these markers, or None:

  • Failed: the tool ran and failed at least once in the period.

  • Denied: a person or a setting refused to let the tool run at least once.

  • New: first seen in the last 30 days.

Hover a row's name to see the raw names Claude reported behind it. Two rows look different from the rest:

  • Connector followed by 8 characters, such as "Connector 0f3a9c21": a connector added in claude.ai that Lemhi can't name. The characters are the start of its id; the full id is in the tooltip.

  • Connectors: connector calls where Claude didn't report which connector was used. Its detail says "Claude didn't report which connector was used. Named connectors are listed by name."

Select a row for its uses, people, failed and denied counts. Tenant Admins also see who used it.

Estimated cost if billed through the API

"What this usage would cost at Anthropic's per-use API prices. It isn't what the client pays for its Claude seats."

Claude estimates a cost for each model call, and Lemhi adds them up. Anthropic describes these values as approximations. Don't use it to reconcile a Claude invoice.

  • Last 30 days: the total.

  • Month over month: the last full month compared with the month before. A dash means the earlier month cost nothing.

  • Weekly, last 12 weeks: cost per week, stacked by model.

  • The split by model: Model, Estimated cost, Share. Under a cent reads <$0.01.

Cost is always for the whole client. It's never broken down by person.

Empty and error states

  • A period with no activity shows "No Claude activity in the last [N] days" (for cost, "in the last 12 weeks"), with the Live or Stopped reporting badge beside Adoption's, so an empty period doesn't read as a broken connection.

  • A view that can't load shows "Couldn't load Claude activity. Try again." with Retry.

Was this helpful?

Still need help? Ask the team