Metermark for Claude Code
Claude Code usage analytics for engineering leaders
Anthropic shows you who ran Claude Code and what it cost. Metermark puts that beside every other AI tool you pay for and shows what the spend is worth: what came out of it, and who’s best at it. Not a word of the conversation stored.
Claude Code activity is not the same as knowing the seat budget works.
Claude Code keeps a good record of itself. Anthropic’s Console shows active users, spend and lines of code accepted per person, and the tool leaves a session history on every machine it runs on. Activity is not the problem.
The problem is the question that follows: is the budget working? Answering it means the Claude Code bill next to Cursor, GitHub Copilot and the API invoice, the work each of them produced, and the people behind both. No vendor dashboard has all three, because each vendor only sees itself.
What Anthropic’s Claude Code admin analytics already cover
Anthropic gives admins a real view of Claude Code, and you should use it.
- The Claude Code analytics dashboard in the Console: active users, sessions, spend, and lines of code accepted with the accept rate, per person and over time.
- The Claude Code analytics API: the same numbers as data, for a warehouse or a spreadsheet.
- OpenTelemetry export: metrics and events from each install into the monitoring stack you already run.
- Plan and spend controls: seats, spend limits and which models a team may use.
What engineering leaders still can’t answer from Claude Code alone
- What the whole AI bill buys. The Console reports Claude Code. The same engineer’s Cursor seat, Copilot licence and Bedrock calls are three other dashboards.
- What kind of work the sessions were. Lines accepted is output. Whether the week went to a launch, a migration or a pile of bug fixes is a different number, and it is the one a project retro needs.
- Cost per outcome. Spend per person is a bill. Spend per shipped project, with the commits and lines changed beside it, is a return.
- Who gets the most from it, and what they do differently. Per-user spend says who is busy. It doesn’t say whose habits are worth writing into the CLAUDE.md everyone else reads.
- Whether to move budget between tools. A seat cut in one place and a bump in another needs both tools measured the same way.
How Metermark fits: a measurement layer, not a Claude Code replacement
Metermark reads the history Claude Code already keeps on each machine. No proxy, no gateway. Each session becomes one row: who, when, which model, what it cost, the kind of work, and a small set of practice scores. The prompt is labelled on the machine and then thrown away; the model’s reply is never opened. Not a word of the conversation stored.
Claude Code sits in the top tier of what Metermark can see, because Anthropic exposes the work. Beside it go the same rows from Cursor, GitHub Copilot, Codex CLI and the rest, each labelled by how much that vendor shows. You keep the Console for Claude Code questions. Metermark answers the cross-tool ones: what every AI tool cost, what came out of it, who’s best at it, and where the AI time on a project went.
A read-only API and an MCP server expose every number, so the answer lands in a spreadsheet, a board deck or a Claude chat rather than another dashboard.
Plugs into
- Claude Code
- Cursor
- GitHub Copilot
- ChatGPT Enterprise
- Claude.ai
- Codex CLI
- Gemini CLI
- AWS Bedrock
- Google Vertex
- Azure OpenAI
- Windsurf
- Devin
Who this page is for
Three people who get asked what the Claude Code seats are worth.
The engineering leader who signs the Claude Code renewal
You want the seat count and the bill defended with something better than “people seem to like it”.
The platform or DevEx team rolling out more than one tool
Claude Code for some, Cursor for others, Copilot by default. You need one way to measure all three.
Finance asking what AI costs per team
One number across tools, split by team, project and kind of work, straight from the server rather than a slide.
Not for reading anyone’s Claude Code conversations. Nobody reads the conversation. Not us, not your boss. Details in the privacy policy.
FAQ
The questions a Claude Code admin asks first.
Does Metermark replace Anthropic’s Claude Code analytics?
No. The Console is the best view of Claude Code on its own, and you should keep using it. Metermark is the layer that puts Claude Code’s cost and work beside every other AI tool you pay for and asks what the spend is worth.
What does Metermark see from Claude Code?
The full picture: who ran the session, when, which model, what it cost, the kind of work it was, and a small set of practice scores. Anthropic exposes the work, which is why Claude Code sits in Metermark’s top coverage tier.
Does Metermark store Claude Code conversations?
No. Not a word of the conversation stored. The prompt is labelled on the machine and deleted, the model’s reply is never opened, and commits and lines changed are counted locally. One row per session leaves the machine.
How does Metermark get Claude Code data?
One command installs the Metermark agent where Claude Code runs, or IT pushes a signed package to the fleet with Jamf, Kandji or Intune. It reads the history Claude Code already keeps and sends a row per session, not the content.
Which other AI tools does Metermark cover?
Claude Code, Cursor, GitHub Copilot, ChatGPT Enterprise, Claude.ai, Codex CLI, Gemini CLI, AWS Bedrock, Google Vertex, Azure OpenAI, Windsurf and Devin, each labelled by how much the vendor exposes.
How much does Metermark cost?
Free for one person, with no card and no contract. Teams pay per seat; the current price is in the pricing section of the home page.
See your own numbers today.
One command. Free for one person. No card, no contract.
Start freeLook at the demo first
Rolling out to a team? Start with yourself, then invite them from the dashboard.