Timebook
io.github.squidcode由 GitHub 帳號 squidcode 發布v0.6.0更新於 Oct 24, 2026
Track time on usetimebook.com - start/stop timers, log entries, list projects/clients.
概覽
Timebook MCP 伺服器,讓助理在 Timebook 帳戶中記錄工時:啟動與停止計時器、登記項目、列出專案與客戶。
- 可以做什麼
- 查看目前登入的使用者,並列出權限範圍內的專案與客戶(唯讀),以便在操作前確認名稱。在專案上啟動與停止執行中的計時器。依時長或依開始與結束時間手動登記工時項目。依專案與日期篩選列出最近的項目。修改項目的欄位或刪除項目,受權杖歸屬與已開立發票項目鎖定限制。
- 適用情境
- 適合希望以對話方式讓 AI 助理在 Timebook 中記錄或調整工時的情境,例如為某項任務啟動計時器、為專案登記工時,或更正與刪除最近的項目。
- 執行需求
- 可使用搭配 OAuth 2.0 的託管 Streamable-HTTP 端點,或透過 stdio 執行本機 npm 套件(需 Node.js 18.17 以上)。本機使用需先在終端機執行一次登入,將 API 權杖存到設定檔。選用環境變數 TIMEBOOK_API_URL 與 TIMEBOOK_WEB_URL 可覆寫 API 與網頁主機位址。
安裝
在 SourceWeft 中
- 開啟 儀表板中的 Timebook,將其新增到工作區。
- 連接 所需的 憑證。它們會被私密儲存,絕不會在公開頁面上展示。
- 為需要使用其工具的對話啟用該服務。
Web executable,透過 Streamable HTTP。 遠端服務在工作區中設定後即可從網頁執行環境執行。
交給代理程式安裝
把這段提示詞貼上到 Claude Code、Codex、Cursor 等代理程式,或 SourceWeft 對話中。代理程式會讀取該伺服器的安裝指南,先向你展示發布者、端點和登入方式,經你同意後再把它新增到你的客戶端。金鑰不會寫進任何命令。
閱讀 https://sourceweft.com/mcp/io.github.squidcode%2Ftimebook/install.md,並按其中說明把這個 MCP 伺服器接入我的客戶端。修改任何設定之前,先告訴我它的發布者、端點和登入方式,等我確認。不要把 API 金鑰寫進命令或我會提交的檔案;問我金鑰應從哪裡取得。其他 MCP 客戶端
把它新增到你客戶端的 mcpServers 設定中。 客戶端會在首次連接時完成 OAuth 登入。
{
"mcpServers": {
"timebook": {
"type": "http",
"url": "https://usetimebook.com/mcp"
}
}
}README
Timebook CLI
Command-line client and MCP server for Timebook — track time, manage timers, and expose your Timebook account to AI agents (Claude, Codex, Cursor, …) over the Model Context Protocol.
[npm] [license] [timebook-cli MCP server]
Install
Requires Node.js 18.17+.
Authenticate
timebook login opens your browser, you log into Timebook (or use an existing session) and pick a scope (which clients/projects this token can touch). The browser delivers the token back to a short-lived loopback HTTP listener, which the CLI then writes to a config file with 0600 permissions.
The token is stored at:
- macOS:
~/Library/Preferences/timebook/config.json - Linux:
~/.config/timebook/config.json - Windows:
%APPDATA%\timebook\Config\config.json
The token never leaves your machine after login. To revoke it server-side, visit https://usetimebook.com/settings/api-tokens.
Use it as a CLI
Duration formats accepted: 1h, 45m, 1h30m, 1.5h, 1:30, or a bare number (interpreted as minutes — e.g. 90 → 1h 30m).
Edit / delete authorization: an API token can only modify entries it created itself. JWT sessions (the web UI) and admin tokens bypass this rule. Invoiced entries are locked for everyone via the API. A 403 with a friendly message is returned on a denied attempt — fix the entry from the web UI or with the token that created it.
Use it as an MCP server
The same binary speaks MCP over stdio when invoked with timebook mcp. Drop it into any MCP-aware host (Claude Code, Claude Desktop, Codex, Cursor, …):
Claude Code / Claude Desktop
Or, if installed globally:
The MCP server reuses the token saved by timebook login — run timebook login once in a terminal before starting the agent.
Use it as a remote MCP (Claude.ai web)
Timebook also runs as a hosted Streamable-HTTP MCP server at https://usetimebook.com/mcp with full OAuth 2.0 (Dynamic Client Registration + PKCE + refresh-token rotation). No CLI install required — Claude.ai discovers it via the standard well-known endpoints:
- Auth-server metadata:
https://usetimebook.com/.well-known/oauth-authorization-server - Resource metadata:
https://usetimebook.com/.well-known/oauth-protected-resource/mcp
Connect from Claude.ai → Settings → Connectors → Add → paste https://usetimebook.com/mcp. You'll be redirected to Timebook's consent page once, then Claude can use all the same tools listed below. Same OAuth-style permissions you'd see for any first-class connector.
The HTTP endpoint also accepts Authorization: Bearer tbk_* (your existing API token) for any client that prefers token-paste over OAuth — including server-to-server use.
Tools exposed to the model
Try it with prompts
Once the MCP server is connected, ask the model in plain English:
- "Start a timer on my Acme website project for landing-page wireframes."
- "How much time did I log on the Recycler project last week?"
- "Log 1 hour 30 minutes against ChatNexus from 9am this morning at the Software Development rate, with description 'code review of the auth refactor'."
- "What am I currently working on?" — invokes
get_active_timer. - "Stop my timer."
- "My last entry on Recycler should be 2 hours, not 1h45m. Fix it." — invokes
list_entriesthenupdate_entry. - "Delete the entry I just made by mistake." — invokes
delete_entry. Will 403 if the entry was created by a different token (web UI, another agent) — say so to the model so it doesn't keep retrying.
The model picks the right tool, asks list_projects first if it needs to disambiguate a name, and writes through start_timer / log_time / stop_timer.
Privacy
Timebook CLI runs on your machine and only talks to your Timebook account.
- Authentication:
timebook loginmints a personal API token via Timebook's OAuth-style consent screen. The token is stored locally with0600permissions (~/Library/Preferences/timebook/config.jsonon macOS,~/.config/timebook/config.jsonon Linux,%APPDATA%\timebook\Config\config.jsonon Windows). It is never transmitted anywhere excepthttps://usetimebook.com(or your override) on outgoing API calls. - Telemetry: none. Neither the CLI nor the MCP server reports usage, errors, or analytics anywhere.
- MCP host data: when you use
timebook mcpfrom inside Claude / Cursor / etc., the MCP host (not Timebook) controls what the model sees. Tool inputs and outputs flow through the host's normal model-context pipeline. - Revoking access: visit https://usetimebook.com/settings/api-tokens to revoke the token at any time.
For Timebook's product-level privacy policy, see https://usetimebook.com/privacy.
Configuration
Override the API/web hosts (useful for self-hosted Timebook or local dev):
You can also pass --api-url and --web-url to timebook login once; subsequent commands re-use the saved values.
If timebook login errors with State mismatch or you want to see exactly which requests reach the loopback callback, run with --debug:
Develop
Pre-commit hooks (ESLint + Prettier via lint-staged) are wired up by husky on npm install.
Release
prepublishOnly runs lint + typecheck + tests + build, then:
License
MIT © Squidcode LLC. See LICENSE.
來源:README.md,提交 d2416f1
工具
0版本歷史
1- v0.6.0最新Sep 16, 2026
評分
已滿足適用檢查項的 85%
該評分衡量此收錄條目的完整程度,依據註冊中心條目、端點與公開來源計算得出。它不是對伺服器本身的評價,也不表示伺服器安全或目前可用:可連通性是帶日期的檢查結果,評分也不包含安裝量或使用者評價。
- 近 30 天內可連通必需25 分
- 已宣告認證方式必需10 分
- 已列出工具15 分
請讓伺服器在無憑證時回應 tools/list,或發布可在沙箱中無憑證啟動的軟體包。
- 描述必需15 分
- 圖示10 分
- 文件15 分
- 授權條款10 分