Let Claude route your next PCB

What it does

KiCAD MCP Server sits between an LLM (Claude, Copilot, etc.) and KiCAD 9.0+, exposing the EDA suite as a set of tools the model can call. You describe a circuit in plain English; the AI translates that into schematic symbols, component placement, copper traces, DRC/ERC checks, and manufacturing exports. It also handles the boring but critical bits: custom footprint generation, personal library management, and real-time JLCPCB part lookup with stock and pricing data.

The interesting bit

The project bundles Freerouting as a Java/Docker auto-router, so the LLM can hand off messy trace optimization rather than pretend it understands differential-pair length matching. Visual snapshots and session logs mean you can audit what the AI actually did to your board—useful when a hallucinated via ruins a power plane.

Key highlights

• Full schematic-to-fabrication workflow via natural language prompts
• Custom symbol/footprint generation for parts outside KiCAD’s standard libraries
• JLCPCB integration with live catalog, pricing, and availability
• Freerouting integration for automatic PCB routing
• Cross-platform: Windows, Linux, macOS
• Built against MCP specification 2025-06-18

Caveats

• Requires KiCAD 9.0+, Node.js 18+, and Python 3.11+—a three-runtime dependency stack
• The disclaimer is unusually explicit: no warranty, AI-generated designs do not replace qualified engineering review, and safety-critical applications require independent expert verification
• “Developed with AI Assistance” (Copilot and Claude) disclosed upfront; code claims human review and integration

Verdict

Hardware hackers who already live in KiCAD and want to accelerate breadboard-to-PCB iteration should try this. Traditional EE layout engineers who trust their muscle memory over an LLM’s guesswork can safely skip it—for now.