print-mcp
Enables generating print-ready PDFs from Markdown and submitting them to CUPS printers. Supports local, Tailscale, and Cloudflare Tunnel access.
README
print-mcp
"We absolutely need to bring ink and paper into this."
print-mcp is a home-network print service that turns Markdown into a print-ready PDF and sends it to your network printer. Run it on any Linux machine on the same home network as the printer, optionally open it to the internet with a Cloudflare Tunnel, and give the MCP endpoint to all of your agents.
Your agent can then:
- Use the network printer configured for your home.
- Print Markdown with page size, margins, orientation, copies, duplex, and color options.
- Check the status of a submitted print job.
The important part: the printer stays on your home network and is never exposed to the internet. Cloudflare only provides the secure public connection to the MCP endpoint; the server renders and submits the job through the network printer.
How It Works
Cloudflare Tunnel
(outbound connection)
|
Any MCP client ------------> MCP server ------------> CUPS ------------> Local printer
| |
Markdown -> PDF USB or IPP
The Docker Compose stack runs three services:
mcpis the MCP server. It renders Markdown and exposes the tools.cupsmanages the local printer queue and submits print jobs.cloudflaredis optional. It publishesmcpthrough a Cloudflare named tunnel without opening an inbound port.
What You Need
- A Linux machine that can reach your home network printer.
- Docker Engine and the Docker Compose plugin.
- A network printer that supports IPP Everywhere, or a USB printer.
- For remote agents: a Cloudflare account, a domain on Cloudflare, and a named tunnel.
Docker Desktop is not supported. USB passthrough and host networking differ from the Linux setup this project expects.
Quick Start
1. Configure the server
git clone <your-repo-url> print-mcp
cd print-mcp
cp .env.example .env
Generate secrets and put them in .env:
openssl rand -hex 32 # MCP_BEARER_TOKEN
openssl rand -hex 24 # CUPS_ADMIN_PASSWORD
At minimum, set these values:
MCP_BEARER_TOKEN=your-long-random-token
CUPS_ADMIN_PASSWORD=your-long-random-password
2. Configure your network printer
For a modern network printer, add its IP address to .env:
CUPS_PRINTER_IP=192.168.1.50
CUPS_PRINTER_NAME=office-printer
When the stack starts, CUPS creates an IPP Everywhere queue at ipp://192.168.1.50/ipp/print, enables it, and makes it the default printer.
If you do not want automatic setup, leave CUPS_PRINTER_IP empty and add a queue manually:
docker compose exec cups lpadmin \
-p office-printer \
-E \
-v ipp://192.168.1.50/ipp/print \
-m everywhere
docker compose exec cups lpadmin -d office-printer
For a USB printer, use the USB Compose overlay instead:
docker compose -f compose.yaml -f compose.usb.yaml up -d --build
3. Start and test locally
docker compose up -d --build
curl -fsS http://127.0.0.1:8000/healthz
The local endpoints are:
- MCP:
http://127.0.0.1:8000/mcp - CUPS admin:
http://127.0.0.1:631/admin
Send a local test print:
echo 'Hello from print-mcp.' | docker compose exec -T mcp \
python /app/cli/print_file.py -
At this point, print-mcp is running on your home network and can print to the configured network printer. Any MCP client on the same network can use the local endpoint; agents outside your network can use the Cloudflare URL from the next section.
Connect A Remote Agent
To let an MCP client connect from anywhere, put the local MCP endpoint behind a Cloudflare Tunnel.
1. Create the Cloudflare route
In Cloudflare Zero Trust:
- Create a named tunnel.
- Add a public hostname, for example
print.example.com. - Set its service URL to exactly
http://mcp:8000. - Copy the tunnel token into
.env.
Then add the same public hostname to the MCP host allow-list:
CLOUDFLARE_TUNNEL_TOKEN=eyJhIjoi...
MCP_ALLOWED_HOSTS=print.example.com
The hostname must match exactly. If it does not, the tunnel can be healthy while the MCP server rejects requests with HTTP 421 Invalid Host header.
2. Start the tunnel
docker compose --profile tunnel up -d --build
Your MCP endpoint is now:
https://print.example.com/mcp
The tunnel makes an outbound connection from your machine to Cloudflare. You do not need to expose port 8000 or open an inbound firewall port.
3. Add it to your MCP client
Configure your MCP-compatible agent or client with the public URL and the same bearer token from .env:
URL: https://print.example.com/mcp
Token: MCP_BEARER_TOKEN
The standard authentication header is:
Authorization: Bearer YOUR_MCP_BEARER_TOKEN
For older MCP agents that cannot set an authorization header, use the same token as the final path segment instead:
https://print.example.com/mcp/YOUR_MCP_BEARER_TOKEN
The server treats this URL as an authenticated alias of /mcp. Path tokens can be
recorded in proxy, tunnel, browser, or server logs, so prefer the header form when
the client supports it.
The exact configuration shape depends on the MCP client. Look for its remote HTTP, Streamable HTTP, or custom MCP server settings.
Now your agent can print.
MCP Tools
The server exposes three tools:
list_printers
Lists configured printers and their current capabilities.
print_markdown
Renders Markdown to PDF and submits it to CUPS. It supports:
titleprinterpage_size:letter,legal, ora4orientation:portraitorlandscapemarginscopies: 1 through 10sides: one-sided, two-sided long-edge, or two-sided short-edgecolor_mode: auto, color, or monochrome
The default is letter paper, portrait orientation, and two-sided long-edge printing.
get_job_status
Returns the current state and timestamps for a CUPS print job.
Raw HTML is disabled. Markdown images may use public HTTP/HTTPS URLs or data URIs. Private, loopback, link-local, multicast, and reserved destinations are blocked, including after redirects.
Local CLI
The same image includes a CLI for printing without an MCP client:
cat notes.md | docker compose exec -T mcp python /app/cli/print_file.py -
docker compose exec -T mcp python /app/cli/print_file.py notes.md \
--page-size a4 \
--orientation portrait \
--margins-mm 12 \
--copies 2 \
--sides two-sided-long-edge \
--color mono
There is also a host-friendly wrapper for files outside the container:
./bin/print-md.sh --page-size a4 ./docs/report.md
Configuration
Copy .env.example to .env for the complete list. The settings most people need are:
| Variable | Purpose |
|---|---|
MCP_BEARER_TOKEN |
Token required by MCP clients |
CUPS_ADMIN_PASSWORD |
Password for CUPS administration |
CUPS_PRINTER_IP |
IP address for automatic IPP queue setup |
CUPS_PRINTER_NAME |
Name of the automatic CUPS queue |
DEFAULT_PRINTER |
Queue used when a tool call omits printer |
CLOUDFLARE_TUNNEL_TOKEN |
Required for the tunnel Compose profile |
MCP_ALLOWED_HOSTS |
Public hostname accepted by the MCP server |
The MCP server and CUPS are bound to localhost by default. Keep them that way when using Cloudflare Tunnel. The tunnel container reaches MCP over the private Compose network.
Troubleshooting
Check service state and logs:
docker compose ps
docker compose logs cups mcp cloudflared
Check the two health endpoints:
curl http://127.0.0.1:8000/healthz
curl http://127.0.0.1:8000/readyz
/healthz confirms that the MCP process is running. /readyz is healthy only after CUPS has at least one configured printer.
Common issues:
- HTTP 421 through Cloudflare: add the exact public hostname to
MCP_ALLOWED_HOSTS, then recreate themcpservice withdocker compose up -d mcp. - Tunnel is running but the endpoint does not respond: verify the Cloudflare public hostname points to
http://mcp:8000and inspectdocker compose logs cloudflared. - Printer is not found: use the printer's explicit IPP URI. Discovery does not reliably cross Docker or Tailscale networks.
- CUPS keeps crashing with
cupsdDoSelect() failed - Bad address!: rebuild the image so the CUPS entrypoint applies the file-descriptor limit workaround.
Development
uv sync --extra dev
uv run ruff check .
uv run pytest
docker compose config
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。