coserial
Enables humans and AI agents to collaboratively control serial ports via MCP tools or Web UI, with real-time data sharing across all participants.
README
coserial
简体中文 | English
Collaborative Serial Debug Bridge. Humans and AI Agents co-control serial ports via MCP tools or Web UI, sharing all IO data in real time. Humans, Agents, and hardware — all on the same page.
Architecture
Agent A ── MCP stdio ── coserial-client ──┐
Agent B ── MCP stdio ── coserial-client ──┤
Agent C ── MCP stdio ── coserial-client ──┤
... │ HTTP (localhost)
Human A ── Web UI ────────────────────────┤
Human B ── Web UI ────────────────────────┤
... ▼
coserial-server
├── HTTP API (/mcp/…) ← MCP tool calls
├── WebSocket /ws ← Web UI real-time data
├── GET / ← Web UI (HTML/CSS/JS)
├── SessionManager
│ ├── COM1
│ ├── COM2
│ └── ...
└── Web UI Monitor
Features
- 🤝 Human-Agent Collaboration — Humans and AI Agents share the same serial session, debugging in real time
- 🔌 MCP Tool Control — Agents read/write serial ports, wait for pattern matches, send commands, etc. via standard MCP protocol
- 🖥️ Web UI Monitor — Dark theme, real-time RX/TX data stream, HEX display, search highlighting, log export
- 🔄 Three-Party Data Sync — Agent I/O and Web UI I/O write to the same buffer — nobody misses anything
- 📡 Multi-Session Management — Connect to multiple serial ports simultaneously, switch via Web UI dropdown
- 🔧 Signal Line Control — DTR/RTS hardware reset support
- 📝 Log Persistence — Serial I/O automatically recorded to file
- 🎯 Zero-Build Frontend — Vanilla HTML/CSS/JS, no frontend toolchain required
Quick Start
Install
# Clone the repo
git clone https://github.com/FlipFlopszzz/coserial.git
cd coserial
# Install dependencies (requires uv)
uv sync
Launch
# Start server + open Web UI (manual debug mode)
uv run coserial
# Start and auto-connect to COM20
uv run coserial COM20
# List all active sessions
uv run coserial list
# Start server only (headless)
uv run coserial-server
MCP Tools
| Tool | Description |
|---|---|
server() |
Discover/start server process, returns port |
open_session(port, baud) |
Create a serial session, returns session_id |
close_session(session_id) |
Close a specific session |
write(session_id, data) |
Send data to serial port |
read(session_id, timeout, size) |
Read buffer data |
wait_for(session_id, pattern, timeout) |
Wait for output matching a regex |
command(session_id, data, expect, timeout) |
Send command and wait for response |
set_params(session_id, baud, ...) |
Dynamically adjust serial parameters |
get_params(session_id) |
Get full session parameters |
set_signals(session_id, dtr, rts) |
Control DTR/RTS signal lines |
list_ports() |
List available serial ports |
list_sessions() |
List all active sessions |
web(port) |
Open Web UI in system browser |
preview(port) |
Return URL for Agent's embedded browser |
shutdown_server() |
Shut down server process |
Web UI
Open http://127.0.0.1:37210/ directly in your system browser, or via any Agent tool's embedded browser.
Project Structure
coserial/
├── pyproject.toml
├── src/coserial/
│ ├── __main__.py # Entry: uv run coserial
│ ├── client.py # MCP stdio client → HTTP proxy to server
│ ├── server.py # Server: HTTP API + WebSocket + Web UI
│ ├── session.py # SessionBuffer + SerialSession + SessionManager
│ └── web_ui/
│ └── index.html # Single-page HTML, inline CSS/JS
Dependencies
- Python >= 3.11
- pyserial — Serial communication
- mcp — MCP SDK (FastMCP)
- websockets — WebSocket server
- aiohttp — HTTP server
- pywin32 — Windows process management
- Icons from IconPark
License
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。