
Timesheet
io.github.theluckystrikev0.22.0更新于 Sep 29, 2026
Timesheet from your AI chat: billable entries, weekly reports, CSV export. Local.
安装
在 SourceWeft 中
- 打开 控制台中的 Timesheet,将其添加到工作区。
- 为需要使用其工具的对话启用该服务。
Desktop only,通过 STDIO。 STDIO 服务会启动本地进程,因此需要 SourceWeft 桌面宿主。
其他 MCP 客户端
参照 仓库 中的启动说明。
README
Track time from Claude with a free, no-install server
Featured on Awesome MCP Servers - directory listing | live hosted endpoint, free tier, no signup.
Track billable time without leaving your AI chat. Say "start a timer on the acme redesign", keep working, then ask for "my hours this week by project" or "invoice lines for acme in August". It keeps a running timer, lets you log time you forgot to track, applies your hourly rate per project, and turns the result into a report, a CSV file or a set of invoice line items. Everything is stored as plain JSON on your own machine.
Built by theluckystrike.
In the official MCP Registry (io.github.theluckystrike/time-tracker-timesheet-billable-hours).
Listed on the AI Product Index - live remote endpoint at mcp.zovo.one/s/time-tracker, free tier, no signup.
Track billable time from chat and turn it straight into a report or invoice line items, zero setup, all local.
60-second install
npm publish for @theluckystrike/mcp-time-tracker is pending. Until then, the .mcpb one-click bundle or a clone+build
is the working path, both are verified below.
One-click (.mcpb): download time-tracker.mcpb from the latest release and double-click it in Claude Desktop:
https://github.com/theluckystrike/mcp-servers/releases/latest
(claude_desktop_config.json):
Claude Code:
(.cursor/mcp.json):
The npx form above starts working the moment the package is published. Until then, use the .mcpb bundle above, or
build from source with exactly these three commands:
Then point your client's command at node with one arg: the absolute path to servers/time-tracker/dist/index.js.
To run in Pro mode set MCP_LICENSE_KEY in the same config block, or call license_activate once with your key.
Tools
Also exposed: the resource timetracker://today (today's summary) and the prompt daily_standup
(writes a standup update from yesterday's and today's tracked time).
What you can say
No tool names required. These are the sentences that were actually tested against the server; the tool column is what answered them.
Two more worth knowing: "export my time to a CSV for my bookkeeper" (export_csv) and "write my standup
update from yesterday and today" (the daily_standup prompt).
Worked example
This is a real transcript from the audit in docs/USER_VALUE_R2.md, numbers unchanged.
One call each. The rate carries its currency all the way through: the report never prints a bare "225", and it never turns into "$225" by accident.
A second worked example, the weekly report and the daily_standup prompt:
Billed hours close
An hour that has been invoiced is finished. entry_mark_billed {ids, invoice_number} writes
billed_at and billed_invoice onto those entries; from then on report and invoice_summary
skip them by default, so next month's "invoice Acme" cannot re-bill work already paid for. The
whole timesheet is still there: pass unbilled_only: false to any of them. invoice_summary
returns the entry_ids it used precisely so they can be handed straight to entry_mark_billed
once the invoice exists.
report and invoice_summary answer overlapping questions on purpose: report is for "how much time
and money," grouped any way you like; invoice_summary is for "give me the lines I can put on an
invoice," which is a narrower, invoice-shaped view of the same entries for one project.
How it stores data
Entries, projects and rates live in one JSON file:
${XDG_DATA_HOME:-~/.local/share}/mcp-servers/time-tracker/data.json.
Every write (starting or stopping a timer, adding, editing or deleting an entry, setting a rate) happens
under an advisory lock file at .../time-tracker/.lock, held across the whole load-mutate-save cycle, so
two overlapping calls cannot interleave and corrupt the file. The save itself writes to a temporary file
and renames it into place, so a crash or a killed process mid-write leaves either the old file or the new
one, never a half-written one. Reads (entry_list, report, timer_status, export_csv) do not take
the lock.
To back up your data, copy the single data.json file (and .lock if present, though it holds no data).
There is no database and no hidden second file.
If data.json is ever unreadable or not valid JSON, the server does not treat that as "no data yet".
It moves the file aside byte-for-byte as data.json.corrupt-<timestamp>, writes a data.json.corrupt
marker and makes every tool, reads included, return data file is corrupt; moved to ...; nothing was written. Restore a good data.json (the quarantined copy is right there) and delete the marker file to
carry on. Nothing is overwritten in the meantime.
Dates, times and rates
- Timestamps with no offset are your local time.
2026-09-02T09:00:00means 09:00 where you are, not UTC. Pass an explicit offset (2026-09-02T09:00:00+02:00) or a trailingZand it is honoured exactly. - Date-only bounds cover whole local days.
from: "2026-09-01"is 00:00:00 local on the 1st andto: "2026-09-30"is 23:59:59.999 local on the 30th, so a month reported by dates includes its last day. Timestamps with a time are used as given. - Entries are clipped to the window. An entry that starts before
fromor ends aftertocounts for the part inside the period, not all of it and not none of it. - Entries are split at local midnight for day grouping. Work from 23:30 to 01:30 is 0.5 h on the first
day and 1.5 h on the next, including across a month boundary.
timer_statuscounts only the part of an entry, or of the running timer, that falls after midnight today. - Rate strings are parsed, never guessed.
"1,200 USD"is 1200 (a comma followed by exactly three digits is thousands grouping),"12,50 EUR"is 12.50 (the unambiguous European decimal shape), and"1.200,50"is 1200.50. Anything that could mean either thing, such as"1,2345", is refused with a worked example instead of being read as the wrong number. - Rates are captured when the time is logged.
entry_addandtimer_stopstore the effective hourly rate and currency on the entry, and reports and invoices use that stored rate.project_set_ratetherefore applies to future entries only; passapply_to_existing: trueto re-rate the time already logged for that project. That re-stamps EVERY entry of the project, including entries that already carry a rate, and the response says how many changed and the project's new total. Addonly_missing: trueto touch only entries that captured no rate of their own. - Tag rows overlap. In
group_by: "tag"an entry taggeddevandmeetingappears in both rows; the total is computed from the entries once, so it is never the sum of the rows.
Limits and honest caveats
- Free
entry_list,report,export_csvandinvoice_summaryonly see the last 7 days. Timers and entries themselves are unlimited and nothing is ever deleted, the window just narrows what a free call can read back. - Free tier supports hourly rates on 2 projects; a third rated project needs Pro.
- Every
reportgrouping is free, tag included: the tag total is a correctness fix, not a premium feature.group_byitself is optional, omit it for the plain total per currency. - Only one timer can run at a time. Starting a second one stops and logs the first, there is no concurrent-timer mode.
- There is no reminder or idle-detection: if you forget to stop a timer, it keeps running until you stop it or start another.
Troubleshooting
npxhangs or fails to find the package: npm publish for this package is pending. Use the.mcpbbundle or the clone-and-build path above until it lands.- Using the
.mcpbbundle: it installs into Claude Desktop directly; there is no separate path to configure. - Using the clone path: the server binary is
servers/time-tracker/dist/index.jsafternpm run build. Point your client'scommandatnodewith that absolute path as the only argument. - Node version: requires Node >= 18. Check with
node -v. - Nothing shows up / silent failures: this server writes logs to stderr only, never stdout (stdout is
reserved for the MCP protocol). In Claude Desktop, check Settings -> Developer -> the server's log
file; in Claude Code, run with
--mcp-debugor check the terminal you launched it from. - A Pro key isn't recognized: run
license_statusto see what the server thinks your tier is, and confirmMCP_LICENSE_KEYis set in the same process the client launches (not just your shell).
Privacy
All data stays local: entries live in ${XDG_DATA_HOME:-~/.local/share}/mcp-servers/time-tracker/data.json.
The server makes no network requests, has no telemetry, and needs no account. License keys are Ed25519
signatures verified offline against a public key compiled into the package, activation works with no
internet connection.
Pairs with
- mcp-invoice, turn
invoice_summaryoutput straight into a numbered PDF invoice. - mcp-spreadsheet, export a CSV with
export_csvand query or reshape it. - mcp-price-tracker, if you also buy things for the client, watch those prices.
- office-suite, every sibling server behind one install, one config entry.
- Guide: Track billable hours in Claude Code and Cursor
FAQ
Yes. All three speak MCP over stdio with the same config shape; the tools and the data file are identical regardless of client.
Nothing is deleted. The entry stays in data.json forever; it just will not appear in entry_list,
report, export_csv or invoice_summary results until you activate Pro, which opens full history.
Yes. Currency is set per project (or per entry, overriding the project default) and every total is grouped by currency, a report never adds EUR and USD together.
The server does not block overlaps; it logs what you tell it. entry_edit lets you fix a mistake after
the fact.
No. There are no network calls anywhere in this server, including for license activation, which is verified with a local public key.
License
MIT
One business profile for the whole suite
Your identity is stored once, at ${XDG_DATA_HOME:-~/.local/share}/mcp-servers/profile/business.json,
and every server in the suite reads it: the invoice issuer, the docx letterhead, the recurring
issuer, expense-tracker's default VAT rate, time-tracker's and timezone's home zone, and the
resume and contract letterheads. Set it once with business_set (invoice or docx) - you never
repeat it anywhere else. An email address is only ever taken from that profile or from an explicit
argument; when none is stored, documents show [add: email] and the tool says so rather than
letting anyone improvise an address.
Frequently asked questions
Is there a free MCP time tracking server?
Yes. The time-tracker server at mcp.zovo.one is a free MCP time tracking server with no install: start and stop timers in chat, keep per-client totals, and produce weekly reports. Unlike SaaS trackers (WebWork, TrackingTime) it needs no account - paste the hosted URL and go.
How do I track time from Claude?
Connect https://mcp.zovo.one/mcp/time-tracker (tokenized URL from mcp.zovo.one/mcp/connect) and say: 'Start a timer for the Acme project.' Stop it later the same way; weekly summaries are one ask away.
Use these docs as an MCP server
Any MCP client (Claude, Cursor, Windsurf, VS Code) can read this repository's documentation directly via GitMCP - no install:
- Docs MCP URL: https://gitmcp.io/theluckystrike/mcp-time-tracker
来源:servers/time-tracker/README.md,提交 e827c57
工具
0版本历史
1- v0.22.0最新Sep 29, 2026

