Time tracking for AI agencies, built to work inside your AI tools.
AI and automation agencies sell outcomes, such as a working support agent or an automated back office, usually for a fixed fee plus a monthly retainer. The hours behind them are hard to see: prompt iteration, evaluation runs, and cleaning up client data. TrackTimer records that time from the tools your team already works in and shows what each build cost.
At a glance
- Claude, ChatGPT, Claude Code, Codex, Cursor, and VS Code can start, switch, and stop timers through TrackTimer’s MCP server when you ask.
- Set each build as a fixed-fee project and see labor cost to date and fee less cost while it is in progress.
- Run post-launch support as a monthly fixed-fee project or an hourly project, depending on the contract.
- On a Mac, AI Auto Track keeps a private timeline of your day, so work between assistant sessions is not lost.
- Contractors see their own hours and earnings; agency margins and billing rates stay private.
- Hours by role on every finished build become the evidence for pricing the next proposal.
Where AI and automation agencies lose time and margin
Fixed fees set before the problem is understood
Discovery shows the client’s data is messier than expected, the first evaluation run fails, and edge cases multiply. The build was priced as a known project; it behaves like research.
Work that happens inside AI tools
Much of the delivery happens in Claude, ChatGPT, Cursor, or a terminal. Leaving that flow to open a separate time tracker is the step people skip, so the hours are rebuilt later from chat history.
Support retainers with blurry edges
After launch come monitoring, prompt fixes, model upgrades, and new requests framed as small changes. Without records, the agency cannot tell whether a retainer covers its cost.
Pricing the next build without history
A young agency has few completed projects to learn from. Every build tracked by role and hour becomes evidence for the next proposal, if the hours were captured.
How TrackTimer works for AI and automation agencies
A fixed fee for the build, a separate project for support
Create a fixed-fee project for each build with the agreed amount. Use a new fixed-fee project each month for a flat support retainer, or an hourly project when support is billed by the hour. Keep discovery calls and internal tooling in non-billable projects.
Timers from the assistant you are already using
Connect Claude, ChatGPT, Claude Code, Codex, Cursor, or VS Code to TrackTimer’s MCP server, sign in, choose a workspace, and approve access. Then ask it to start a billable timer with a note, switch to another client, stop, show your day, or correct a completed entry with a reason. The assistant acts only when asked and only on your own time.
AI Auto Track for everything else
On a Mac, AI Auto Track keeps a private timeline of your day. Each block gets a title and summary, and any block can become a time entry with its summary as the description. Teammates and clients never see the timeline.
Cost each build by role
Each person has a pay rate and billing rate, with overrides by client or project. The fixed-fee project shows the agreed fee, tracked time, labor cost to date, and fee less cost, including running timers.
Contractors without access to your margins
Invite freelance engineers as contractors. They track only the projects they are assigned to and see their own hours and earnings, never client billing rates or agency financials.
Invoice fees and hours in Mercury
With Mercury connected, select a build’s agreed fee or hourly support time at saved rates, and create an unsent invoice in Mercury. Review and send it from Mercury.
Example: a $25,000 customer-support agent
A three-person agency builds a customer-support agent for a fixed fee of $25,000, including integration with the client’s help desk and an evaluation suite. A freelance automation engineer handles the integrations. At launch:
| Person | Tracked | Pay rate | Labor cost |
|---|---|---|---|
| Lead AI engineer | 60h | $100/h | $6,000 |
| Automation engineer (contractor) | 90h | $80/h | $7,200 |
| Solutions consultant | 30h | $70/h | $2,100 |
| Total | 180h | $15,300 |
- Agreed fee
- $25,000
- Tracked time
- 180h
- Labor cost
- $15,300
- Fee less cost
- $9,700 · 39%
Illustrative example with hypothetical people, hours, and rates. The build brought in $138.89 per tracked hour ($25,000 ÷ 180h), and the next proposal can start from 180 hours and a known mix of roles instead of a guess.
Set up your first week
Add clients and projects. Create a fixed-fee project for each build and a project for each support arrangement, with the right billing mode.
Set billing and pay rates. Give each person defaults and override them for contractors or clients with different terms.
Invite the team. Invite employees as admins or contractors, and assign contractors to the projects they work on.
Connect your AI tools. Follow the MCP guide to connect each person’s assistant. Each connection is authorized in the browser and bound to one workspace.
Review each build. Check fee less cost weekly, then use finished builds’ hours by role to price the next proposal.
Frequently asked questions
Can Claude or ChatGPT start and stop my timer?
Yes. After you connect TrackTimer’s MCP server and approve access, you can ask your assistant to start a timer on a project with a note, switch projects, stop, review your entries, or correct a completed entry with a reason.
How does automatic tracking fit with AI tools?
Your assistant acts when you ask, so you stay in control of what is recorded. For automatic capture on a Mac, AI Auto Track keeps a private timeline of your day that you can turn into time entries.
What can a connected assistant see?
Only the connected person’s own time in the workspace they approved: their clients, projects, entries, audit history, and earned pay. Access stays that focused even for owners and admins, which keeps teammates’ time and agency financials private.
Can contractors connect their own AI tools?
Yes. Each person authorizes their own connection in the browser for one workspace, and it covers only their own time and earnings.
How should we price a support retainer?
One approach is to track the first months of support as an hourly project, then set a flat monthly fee from the actual hours. After that, use a fixed-fee project each month to see whether the retainer still covers its labor cost.
Is there an API for our own automations?
Yes. The REST API uses personal keys bound to one workspace to list clients and projects, start and stop timers, and read your time entries. Owners and admins can create projects.
See what your time earns.
Set up your first client and project in a few minutes. Free during early access.
Written by TrackTimer about its own product and reviewed . Examples use hypothetical people, hours, and rates. Features and planned pricing can change; the pricing section has the current terms.