timing-cli
Provides MCP tools to read local Timing.app activity data, generate aggregated time entries, and push them to the Timing Web API.
README
timing-cli
A command-line interface and MCP server for Timing.app
on macOS. Unlike the Timing Web API — which cannot see your locally recorded app
usage — timing-cli reads Timing's local activity database directly
(read-only) and turns real app usage into aggregated time entries, which it can
then push back to Timing via the Web API.
Because it ships an embedded MCP server, agents (e.g. the Hermes personal agent)
can query your activity and create entries without ever copying or exposing
the raw SQLite.db — the server runs locally, next to the database.
Not affiliated with Timing. Independent tool. It only ever reads the local database; all writes go through the official Web API.
What it does
- Reads the local Timing store (
~/Library/Application Support/info.eurocomp.Timing2/SQLite.db) read-only: automatic app activity, window titles, document paths, projects. - Classifies unassigned activity onto projects with your own rules (Timing's built-in predicate rules only cover ~15% of activity). It also reuses decoded Timing project predicate rules from the local database when Timing did not already assign a slice.
- Aggregates consecutive same-project slices into clean time blocks (gap-merging + minimum-duration filtering).
- Pushes the resulting suggestions to Timing as real time entries via the Web API (with a safe dry-run default).
- Serves all of this over MCP (
timing serve) for agents.
Quickstart
# Install (globally, via uv)
uv tool install timing-cli
# See what's in your local database
timing info
# Daily project summary and suggested entries (read-only)
timing summary --date 2026-07-05
timing suggest --date 2026-07-05
# Push suggestions to Timing (dry-run first, then --yes)
export TIMING_API_KEY=... # from https://web.timingapp.com/integrations/tokens
timing push --date 2026-07-05 # dry-run
timing push --date 2026-07-05 --yes # actually create entries
# Run the MCP server (for Hermes / other agents)
timing serve # stdio
export TIMING_MCP_TOKEN=...
timing serve --transport http # HTTP on 127.0.0.1:8321
timing push --yes resolves every non-unassigned suggestion to a unique Web-API
project before creating anything. If a project is ambiguous or unmapped, the push
fails before the first write; add a [project_mappings] override. Re-running the
same push skips matching existing entries unless --replace is passed.
Commands
| Command | Description |
|---|---|
timing info |
Database location, recorded date range, token status |
timing projects [--remote] [--archived] |
List projects (local DB or Web API) |
timing usage [--date/--from/--to] [--project ID] |
Raw automatically tracked app usage |
timing summary [--date/--from/--to] |
Total time per project |
timing suggest [--date/--from/--to] |
Aggregated time-entry suggestions (read-only) |
timing push [--date/--from/--to] [--yes] [--replace] |
Create entries via Web API (dry-run by default) |
timing serve [--transport] [--host] [--port] |
Run the MCP server |
Daily workflow
# 1. Inspect the day without writing anything.
timing summary --date 2026-07-05
timing suggest --date 2026-07-05
# 2. If push projects are unmapped, inspect remote project references.
export TIMING_API_KEY=...
timing projects --remote
# 3. Add missing [project_mappings] entries, then dry-run again.
timing push --date 2026-07-05
# 4. Create entries once the dry-run looks right.
timing push --date 2026-07-05 --yes
Re-running step 4 for the same day skips matching existing entries. Use
--replace only when you deliberately want Timing's API to replace overlapping
entries in the target window.
Configuration
Optional config at ~/.config/timing-cli/config.toml:
# Override the database path if Timing lives elsewhere.
# db_path = "~/Library/Application Support/info.eurocomp.Timing2/SQLite.db"
api_base_url = "https://web.timingapp.com/api/v1"
# api_token = "..." # prefer the TIMING_API_KEY env var instead
# mcp_http_token = "..." # prefer the TIMING_MCP_TOKEN env var instead
min_block_seconds = 120 # drop aggregated blocks shorter than this
gap_merge_seconds = 300 # merge same-project slices split by a gap up to this
# Optional overrides from local Timing projects to Web-API project references.
# Keys can be a local id ("id:42"), a full title chain, or a leaf title.
[project_mappings]
"Client / Polaris" = "/projects/123"
"id:42" = "/projects/456"
# Classification rules: map unassigned activity onto projects.
# First match wins. `app`/`bundle_id` are case-insensitive substrings;
# `title`/`path` are regexes.
[[rules]]
project = "Polaris"
title = "polaris"
[[rules]]
project = "Cognovis"
path = "code/mira"
Cognovis example
This is a real-world shape for Malte's local setup. Fill the Web-API project
references from timing projects --remote; local project ids can be discovered
with timing projects.
api_base_url = "https://web.timingapp.com/api/v1"
min_block_seconds = 120
gap_merge_seconds = 300
[project_mappings]
"cognovis Verwaltung" = "/projects/REMOTE_COGNOVIS_VERWALTUNG"
"]project-open[" = "/projects/REMOTE_PROJECT_OPEN"
"Home Electronic" = "/projects/REMOTE_HOME_ELECTRONIC"
[[rules]]
project = "cognovis Verwaltung"
title = "(timing|collmex|paperless|invoice|rechnung)"
[[rules]]
project = "cognovis Verwaltung"
path = "code/(cli-tools|library)"
[[rules]]
project = "]project-open["
app = "Mail"
title = "project-open"
Timing's own project predicates are loaded from the local Project.predicate
column automatically and applied after explicit config rules. Keep local rules
for repository paths, editor titles, and project-specific conventions that
Timing itself does not classify well.
MCP tools
timing serve exposes: list_timing_projects, list_app_usage_tool,
daily_project_summary, suggest_time_entries, create_time_entry (write),
recorded_date_range.
HTTP transport requires bearer-token authentication via TIMING_MCP_TOKEN or
mcp_http_token in the config. Stdio transport remains local and does not require
an MCP token.
Release
The repository includes a GitHub Actions release workflow at
.github/workflows/release.yml. Tag pushes run tests and lint only; published
GitHub releases or manual dispatches build the package and publish to PyPI using
Trusted Publishing. Configure a PyPI trusted publisher for the repository and
the pypi environment before running the publish job.
Requirements
- macOS with Timing.app installed
- Python 3.12+ (installed automatically by
uv tool install) - A Timing Web API token for pushing entries (read-only commands need no token)
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。