CaffeineCurfew
Tracks caffeine decay and predicts safe bedtimes using a half-life model. No external integrations needed, intake is passed as parameters.
README
Caffeine Curfew MCP Server
A local Model Context Protocol server that tracks caffeine decay and predicts safe bedtimes. All caffeine intake is passed directly as parameters. No external integrations are required.
How It Works
Caffeine decays according to the standard half-life formula:
remaining = initial * (0.5 ^ (hours_elapsed / half_life))
The server calculates the sum of that formula across every entry in your intake log to determine your total current caffeine level and find the earliest time it drops below your personal sleep interference threshold.
The half-life is adjustable from 3 to 10 hours (default 5). The sleep interference threshold is adjustable from 10 to 50 mg (default 25 mg).
Quickstart for Claude Code Users
No installation required. Add this block to your Claude Code MCP settings file and restart Claude Code. uvx will fetch and run the server automatically on first use.
On macOS the settings file is located at:
~/Library/Application Support/Claude/claude_desktop_config.json
{
"mcpServers": {
"caffeineCurfew": {
"command": "uvx",
"args": ["caffeine-curfew-mcp"]
}
}
}
That is the only step. The four Caffeine Curfew tools will appear in Claude Code after restart.
Tools
get_caffeine_level
Returns the current total caffeine level in the body.
Parameters:
entries list of {amount_mg, consumed_at} intake records
half_life_hours float, caffeine half-life in hours (default 5)
get_safe_bedtime
Returns the earliest time at which caffeine will drop below the threshold.
Parameters:
entries list of {amount_mg, consumed_at} intake records
half_life_hours float (default 5)
threshold_mg float, sleep interference threshold in mg (default 25)
simulate_drink
Shows how adding a new drink right now would shift the safe bedtime.
Parameters:
entries list of {amount_mg, consumed_at} intake records
half_life_hours float (default 5)
threshold_mg float (default 25)
new_drink_mg float, caffeine content of the hypothetical drink in mg
get_status_summary
Returns a full summary including current level, safe bedtime, and whether a target bedtime is reachable.
Parameters:
entries list of {amount_mg, consumed_at} intake records
half_life_hours float (default 5)
threshold_mg float (default 25)
target_bedtime string, optional ISO 8601 datetime the user wants to sleep by
Sample Entries for Testing
The timestamps below should be replaced with real ISO 8601 datetimes that are a few hours before the current time when you run the tests.
[
{
"amount_mg": 150,
"consumed_at": "2025-01-15T08:00:00"
},
{
"amount_mg": 80,
"consumed_at": "2025-01-15T12:30:00"
},
{
"amount_mg": 100,
"consumed_at": "2025-01-15T15:00:00"
}
]
The first entry represents a morning coffee (approximately 150 mg). The second represents a lunchtime espresso (approximately 80 mg). The third represents an afternoon cold brew (approximately 100 mg).
A typical get_status_summary call with the sample entries and a 10 pm target bedtime:
{
"entries": [
{"amount_mg": 150, "consumed_at": "2025-01-15T08:00:00"},
{"amount_mg": 80, "consumed_at": "2025-01-15T12:30:00"},
{"amount_mg": 100, "consumed_at": "2025-01-15T15:00:00"}
],
"half_life_hours": 5,
"threshold_mg": 25,
"target_bedtime": "2025-01-15T22:00:00"
}
Remote Deployment (Claude Mobile App)
To use Caffeine Curfew from the Claude iOS or Android app, you need the server running on a machine that is reachable over the internet with a valid HTTPS URL. A spare Linux machine on your home network works well for this.
The recommended setup uses Cloudflare Tunnel, which gives you a free persistent HTTPS URL without opening any ports on your router.
Step 1: Install the server on Linux
pip3 install caffeine-curfew-mcp
Or clone the repo and install in editable mode for easier updates:
git clone https://github.com/YOUR_USERNAME/caffeine-curfew-mcp
cd caffeine-curfew-mcp
pip3 install -e .
Step 2: Install Cloudflare Tunnel
curl -L https://github.com/cloudflare/cloudflared/releases/latest/download/cloudflared-linux-amd64.deb -o cloudflared.deb
sudo dpkg -i cloudflared.deb
Then log in to your Cloudflare account (free account at cloudflare.com is enough):
cloudflared tunnel login
Create a named tunnel:
cloudflared tunnel create caffeine-curfew
Note the tunnel ID that is printed. Create a config file at
~/.cloudflared/config.yml:
tunnel: YOUR_TUNNEL_ID
credentials-file: /home/YOUR_USER/.cloudflared/YOUR_TUNNEL_ID.json
ingress:
- hostname: caffeine.yourdomain.com
service: http://localhost:8000
- service: http_status:404
Route your domain to the tunnel (requires a domain managed by Cloudflare):
cloudflared tunnel route dns caffeine-curfew caffeine.yourdomain.com
If you do not have a domain, you can use a temporary public URL for testing:
cloudflared tunnel --url http://localhost:8000
This prints a random URL like https://random-words.trycloudflare.com. It works
immediately but changes every time the process restarts.
Step 3: Start the server
In one terminal, start the MCP server in SSE mode:
caffeine-curfew-mcp --transport sse
In another terminal, start the Cloudflare Tunnel:
cloudflared tunnel run caffeine-curfew
To run both automatically on boot, create systemd services for each. A minimal
service file for the MCP server at /etc/systemd/system/caffeine-curfew.service:
[Unit]
Description=Caffeine Curfew MCP Server
After=network.target
[Service]
ExecStart=/usr/local/bin/caffeine-curfew-mcp --transport sse
Restart=always
User=YOUR_USER
[Install]
WantedBy=multi-user.target
Enable and start it:
sudo systemctl enable caffeine-curfew
sudo systemctl start caffeine-curfew
Step 4: Connect in the Claude App
Open the Claude app, go to Settings, then Integrations (or Tools, depending on app version), and add a new MCP server with the URL:
https://caffeine.yourdomain.com/sse
The four Caffeine Curfew tools will be available in every conversation.
Security note
Anyone with the URL can call your server. If you want to restrict access, add a shared secret by setting an environment variable before starting the server and putting the Cloudflare Tunnel behind Cloudflare Access (free for personal use), which adds an authentication layer without changing the server code.
Local Development Setup
Clone the repository and install in editable mode:
git clone https://github.com/YOUR_USERNAME/caffeine-curfew-mcp
cd caffeine-curfew-mcp
pip3 install -e ".[dev]"
Run the server directly:
python3 -m caffeine_curfew.server
To point your local Claude Code at a development checkout instead of the published package, use the python3 form in your MCP settings:
{
"mcpServers": {
"caffeineCurfew": {
"command": "python3",
"args": ["/absolute/path/to/caffeine-curfew-mcp/caffeine_curfew/server.py"]
}
}
}
Publishing to PyPI
Build and upload with standard tooling:
pip3 install build twine
python3 -m build
python3 -m twine upload dist/*
After publishing, any user with uvx installed can connect to the server using the one-line config shown in the Quickstart section above.
Notes
Caffeine half-life varies significantly between individuals. Common ranges:
Average adult 4 to 6 hours
Oral contraceptives can double the half-life
Smokers roughly half the half-life of non-smokers
Pregnancy can extend half-life to 15 hours or more
The default half-life of 5 hours is a reasonable starting point for most adults. All times returned are provided in both UTC and the local system timezone.
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。