Organic Open-Source Plugin + MCP EDA · v1.6.32
From your first prompt to production-ready GDS — fully AI-driven. No IC design experience required.
Every IC starts from a user-friendly dialogue and ends with a verified chip on FPGA.
No technical jargon — AI translates your product needs into silicon.
What users can understand
What engineers can understand
What the foundry needs
An organic plugin built on the world's strongest AI. General skills grow as the AI giant grows. Domain expertise deepens through community contributions.
General skills — automatic
Domain expertise — community-driven
Open to everyone — plugin your device, your IP, your knowledge.
FPGA boards, testers, scopes, lab equipment
manifest.json in src/devices/ — your device becomes an MCP tool instantlyReusable IP blocks for the community
Community backlogs that evolve the plugin
Model Context Protocol is the wire between Claude and the silicon stack. We open-sourced an MCP server that exposes 46 tools — synthesis, place & route, GDS export, FPGA programming, scope capture, protocol testing, SPICE corner sweep, analog layout, xschem netlist, ADC read, oracle dump — as one declarative interface.
A protocol, not a framework
mcp-eda-server v0.114.0 — 46 tools
Self-healing diagnostics + anti-fabrication
Each skill is a domain expert guiding one step of the IC design flow.
Layer 1: compliance.yaml regex checks agent output text.
Layer 2: programs verify actual artifacts.
Layer 3: MCP execution proof + provenance hash match.
Layer 4: real-hardware pass attestation (L13).
Open-source tools, open PDKs, open AI — no vendor lock-in.
Powered by IIC-OSIC-TOOLS — 37 MCP EDA tools in one Docker image
GF180MCU + SKY130 (open) + custom PDK support
Intel MAX 10 (DE10-Lite) + Quartus Prime Lite — RTL prototype before silicon
Claude Code + MCP Server v0.114.0 — 46 tools (37 EDA + 9 device), custom PDK
1902 tests, 350 programs, 5-rule anti-fabrication doctrine, 213-gate checklist
PostgreSQL + pgvector — IC spec semantic search
Lower the barrier from decades of training to a conversation.
Four steps. The Docker image (~22 GB) bundles every EDA tool. Hardware lab is optional — simulation and GDS tapeout flows do not need physical devices.
docker pull hpretl/iic-osic-tools:latest
docker run -d --name iic-eda \
-v $HOME/designs:/design \
hpretl/iic-osic-tools:latest --wait
--wait keeps the container alive so MCP can docker exec into it.
git clone https://github.com/reyerchu/\
AI_IC_design.git
cd AI_IC_design/mcp-eda-server
npm install
claude mcp add eda-tools \
node $(pwd)/src/index.js \
-e EDA_CONTAINER=iic-eda
Verify with claude mcp list — eda-tools should be listed.
# start an interactive Claude session
claude
# inside the session, run two slash commands:
> /plugin marketplace add \
/path/to/AI_IC_design/vibe-ic-marketplace
> /plugin install vibe-ic@vibe-ic-marketplace
/plugin commands run inside Claude Code — not in your shell.
claude "Design a temperature sensor IC
with I2C interface, 12-bit,
alert output, SOIC-8 package"
Or invoke directly: /vibe-ic-all
Need full troubleshooting? See mcp-eda-server/INSTALL_GUIDE.md for prerequisites, Docker / Node / Quartus setup, and step-by-step verification.
Physical devices for FPGA bring-up and on-board verification. Not required for simulation or GDS tapeout flows.
# 1. Install Quartus 23.1 Lite (free)
# intel.com/Quartus → Lite Edition
# 2. USB-Blaster udev rule
sudo cp quartus/linux64/51-usbblaster.rules \
/etc/udev/rules.d/ && sudo udevadm control --reload
# 3. Tell MCP server where Quartus lives
# Edit .claude/.mcp.json → env:
"QUARTUS_ROOTDIR": "/path/to/quartus",
"PATH": "/path/to/quartus/bin:..."
# 4. Verify: restart claude, then run
# eda_device_list → quartus_pgm: satisfied
# No driver needed (Linux USB TMC built-in)
# 1. Plug in scope via USB
# 2. Grant permission
sudo usermod -aG plugdev $USER
# 3. Verify: log out/in, then
lsusb | grep 2a8d
# 4. Restart claude → eda_device_list
# → scope: plugdev satisfied
Want to ship your own device, IP, or partner plugin? See the Open Platform page.