kicad-mcp
An MCP server that gives AI agents end-to-end control of KiCad 9+ for rule checks, manufacturing exports, production-readiness certification, and live PCB editor control.
README
kicad-mcp
An MCP server that gives an AI agent end-to-end control of KiCad 9+: rule checks, manufacturing exports, a production-readiness certification gate, live PCB-editor control, and netlist generation.
It is built around a simple split that mirrors how KiCad actually works:
kicad-cli(headless, no running instance) handles DRC/ERC and every export (Gerbers, drill, position, BOM, STEP, PDF, SVG, netlist, PNG render).kipy(the KiCad 9 IPC API) handles live control of a board open in the PCB editor.- A built-in s-expression parser reads
.kicad_pcb/.kicad_schfiles offline with zero dependencies, so inspection always works. - SKiDL turns Python hardware descriptions into importable netlists.
kicad_preflightties the checks and exports together into a single pass/fail certification that only emits a fabrication package when the board actually passes.
Surfaces and tools
| Surface | Tools |
|---|---|
Headless checks & exports (kicad-cli) |
kicad_cli_info, kicad_run_drc, kicad_run_erc, kicad_export, kicad_generate_fab_package |
| Certification workflow | kicad_preflight |
| Offline inspection & safe edits | kicad_inspect_project, kicad_set_title_block, kicad_set_symbol_field |
Live IPC control (kipy) |
kicad_connect, kicad_board_summary, kicad_list_footprints, kicad_move_footprint, kicad_set_footprint_value, kicad_save_board |
| Netlist generation (SKiDL) | kicad_generate_netlist |
kicad_preflight is the headline tool. Given a board (and ideally its schematic) it runs ERC, DRC with schematic parity, a fully-routed check, and a board-outline check as hard gates, plus soft checks for missing footprint values and unplaced parts. Only when every hard gate passes does it generate the Gerber/drill/position/BOM package (zipped), a STEP model, and PCB/schematic PDFs.
Installation
Requires Python 3.10+ and KiCad 9.0+.
pip install -e .[all]
The base install depends only on mcp and pydantic. The optional bindings are pulled in by extras and each degrades gracefully — if one is missing, only its tools report an actionable install hint, and the rest of the server keeps working:
.[live]→kicad-python(live IPC control).[edit]→kiutils(offline field edits).[netlist]→skidl(netlist generation).[all]→ everything
kicad-cli itself ships with KiCad; it is not a pip package. The server finds it automatically via the KICAD_CLI environment variable, then PATH, then the standard install locations:
- Windows:
C:\Program Files\KiCad\9.0\bin\kicad-cli.exe - macOS:
/Applications/KiCad/KiCad.app/Contents/MacOS/kicad-cli - Linux:
/usr/bin/kicad-cli
If it lives elsewhere, set KICAD_CLI to the full path.
Enabling live control
The live tools talk to a running KiCad over its IPC API. In KiCad, open Preferences → Plugins and enable the KiCad API, then keep a board open in the PCB editor. The IPC API in KiCad 9/10 covers the PCB editor only and does not do exports — that is why exports go through kicad-cli.
Claude Desktop configuration
Add the server to claude_desktop_config.json:
{
"mcpServers": {
"kicad": {
"command": "python",
"args": ["-m", "kicad_mcp"],
"env": {
"KICAD_CLI": "C:\\Program Files\\KiCad\\9.0\\bin\\kicad-cli.exe"
}
}
}
}
On Windows, the config file location depends on how Claude Desktop was installed:
- Standard install:
%APPDATA%\Claude\claude_desktop_config.json - MSIX / Microsoft Store install:
%APPDATA%is redirected into the package sandbox, so the file lives underC:\Users\<you>\AppData\Local\Packages\<ClaudePackageFamilyName>\LocalCache\Roaming\Claude\claude_desktop_config.json
The reliable way to avoid guessing is to open the file from inside the app (Settings → Developer → Edit Config), which always writes to the correct location for your install. Point command at the Python interpreter of the environment where you installed kicad-mcp (a full path to python.exe avoids PATH surprises under MSIX). A sample config is in examples/claude_desktop_config.json.
Notes
- Tool inputs are validated with Pydantic; malformed calls fail fast with a clear message.
- Rule-check tools parse
kicad-cli's JSON reports, so violations come back as structured severity counts rather than scraped text. - Live edits (
kicad_move_footprint,kicad_set_footprint_value) apply to the open editor immediately but are not written to disk untilkicad_save_board. kicad_inspect_projectnever needs KiCad installed at all.
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。