EDA-MCP-Server by dukeceicenter - MCP Server

MCP EDA - Model Context Protocol for Electronic Design Automation

A comprehensive Model Context Protocol (MCP) based Electronic Design Automation (EDA) system that provides intelligent automation for digital design implementation flows. This system integrates multiple EDA tools through a unified MCP interface, enabling natural language-driven design automation with experimental evaluation framework.

Features

AI-Powered Automation: Natural language interface for EDA operations using GPT-4
Multi-Tool Integration: Seamless integration of Synopsys Design Compiler and Cadence Innovus
Complete Design Flow: End-to-end digital design implementation from RTL to GDSII
RESTful API: Standardized HTTP API for all EDA operations
Real-time Monitoring: Live status tracking and detailed logging
Modular Architecture: Independent microservices for each design stage
Configuration Management: CSV-based parameter management for design optimization
Experimental Framework: Comprehensive TCL accuracy evaluation system
Multiple Design Support: Support for various benchmark designs (des, b14, leon2, etc.)

System Architecture

┌─────────────────┐    ┌─────────────────┐    ┌─────────────────┐
│   User Query    │───▶│  MCP Agent      │───▶│  MCP Servers    │
│  (Natural Lang) │    │  Client (8000)  │    │  (13333-13440)  │
└─────────────────┘    └─────────────────┘    └─────────────────┘
                                │                       │
                                ▼                       ▼
                       ┌─────────────────┐    ┌─────────────────┐
                       │   OpenAI GPT-4  │    │   EDA Tools     │
                       │   (Tool Select) │    │  (DC/Innovus)   │
                       └─────────────────┘    └─────────────────┘

MCP Server Architecture

Service	Port	Endpoint	Description
MCP Agent Client	8000	`/agent`	Main interface for natural language processing
Synthesis Setup	13333	`/setup/run`	RTL synthesis setup and configuration
Synthesis Compile	13334	`/compile/run`	RTL-to-gate synthesis compilation
Floorplan	13335	`/floorplan/run`	Chip floorplanning and I/O placement
Power Planning	13336	`/power/run`	Power distribution network generation
Placement	13337	`/place/run`	Standard cell placement optimization
Clock Tree Synthesis	13338	`/cts/run`	Clock distribution network synthesis
Routing	13339	`/route/run`	Signal routing and optimization
Save Design	13440	`/save/run`	Final design saving and output generation

Experimental Framework

The project includes a comprehensive experimental framework for evaluating TCL generation accuracy across different methods:

Experiment Methods

Baseline1: Pure LLM-based TCL generation
Baseline2: LLM + template-based generation
Ours: MCP agent with real EDA tool execution

Experiment Structure

experiment/
├── generate/                    # TCL generation scripts
│   ├── baseline1_generate.py   # Pure LLM generation
│   ├── baseline2_generate.py   # LLM + template generation
│   ├── ours_generate.py        # MCP agent generation
│   └── common_config.py        # Shared configuration
├── evaluate/                    # Evaluation scripts
│   ├── tcl_evaluator.py        # Main evaluator
│   └── evaluation_metrics.py   # Quality metrics
├── results/                     # Generation results
├── evaluation_results/          # Evaluation reports
└── run_experiment.py           # Main experiment runner

Running Experiments

# Run complete experiment (generate + evaluate)
cd experiment
python run_experiment.py --full

# Run only generation for specific methods
python run_experiment.py --generate baseline1 ours

# Run only evaluation on existing results
python run_experiment.py --evaluate

# Clean previous results
python run_experiment.py --clean --full

Prerequisites

Software Requirements

Python 3.9+
Synopsys Design Compiler (for synthesis)
Cadence Innovus (for physical implementation)
FreePDK45 technology library
OpenAI API Key (for GPT-4 integration)

System Requirements

OS: Linux (tested on Ubuntu 20.04+)
Memory: 8GB RAM minimum (16GB+ recommended)
Storage: 10GB free space for design files and logs
Network: Local network access for inter-service communication

Installation

Option 1: Docker Deployment (Recommended)

Due to licensing restrictions of commercial EDA tools, we use a hybrid Docker approach:

# Clone the repository
git clone https://github.com/AndyLu666/MCP-EDA-Server.git
cd MCP-EDA-Server

# Copy environment template
cp docker/env.example .env

# Edit .env file with your configuration
# - Set your OpenAI API key
# - Configure EDA tools host
# - Set license file paths

# Deploy using Docker
chmod +x docker/deploy.sh
./docker/deploy.sh deploy

Note: EDA tools (Design Compiler, Innovus) must be installed on the host system with valid licenses. The Docker containers communicate with these tools via network.

Option 2: Local Installation

1. Clone the Repository

git clone https://github.com/AndyLu666/MCP-EDA-Server.git
cd MCP-EDA-Server

2. Set Up Python Environment

# Create virtual environment
python3 -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

3. Configure Environment Variables

Create a .env file in the project root:

# OpenAI API Configuration
OPENAI_API_KEY=your_openai_api_key_here

# MCP Server Configuration
MCP_SERVER_HOST=http://localhost

# Optional: Custom ports (defaults shown)
SETUP_PORT=13333
COMPILE_PORT=13334
FLOORPLAN_PORT=13335
POWER_PORT=13336
PLACE_PORT=13337
CTS_PORT=13338
ROUTE_PORT=13339
SAVE_PORT=13440

4. Verify EDA Tools Installation

# Check Design Compiler
which dc_shell

# Check Innovus
which innovus

# Verify FreePDK45 library
ls libraries/FreePDK45/

Quick Start

1. Start MCP Agent Client

# Terminal 1: Start the main agent
conda activate eda310  # or your conda environment
uvicorn mcp_agent_client:app --reload --host 0.0.0.0 --port 8000

2. Start MCP Servers

# Terminal 2: Start all MCP servers
./restart_servers.sh

# Or start individually:
cd server
python3 synth_setup_server.py --port 13333 &
python3 synth_compile_server.py --port 13334 &
python3 floorplan_server.py --port 13335 &
python3 powerplan_server.py --port 13336 &
python3 placement_server.py --port 13337 &
python3 cts_server.py --port 13338 &
python3 route_server.py --port 13339 &
python3 save_server.py --port 13440 &

3. Run Your First Design

# Test with the example design
curl -X POST http://localhost:8000/agent \
  -H "Content-Type: application/json" \
  -d '{"user_query":"Run synth_setup for design=\"des\" and return the log path."}'

Usage Guide

Natural Language Interface

The MCP Agent Client understands natural language queries and automatically selects the appropriate EDA tool:

# Synthesis
curl -X POST http://localhost:8000/agent \
  -d '{"user_query":"Run synth_setup for design=\"my_design\" and return the log path."}'

curl -X POST http://localhost:8000/agent \
  -d '{"user_query":"Run synth_compile for design=\"my_design\" and return the log path."}'

# Physical Implementation
curl -X POST http://localhost:8000/agent \
  -d '{"user_query":"Run floorplan for design=\"my_design\", top_module=\"my_module\" and return the path to floorplan.enc.dat."}'

curl -X POST http://localhost:8000/agent \
  -d '{"user_query":"Run placement for design=\"my_design\", top_module=\"my_module\" and return the path to placement.enc.dat."}'

curl -X POST http://localhost:8000/agent \
  -d '{"user_query":"Run routing for design=\"my_design\", top_module=\"my_module\" and return the path to route_opt.enc.dat."}'

curl -X POST http://localhost:8000/agent \
  -d '{"user_query":"Run save for design=\"my_design\", top_module=\"my_module\" and return the archive path."}'

Complete Design Flow Example

# 1. Synthesis Setup
curl -X POST http://localhost:8000/agent \
  -d '{"user_query":"Run synth_setup for design=\"des\" and return the log path."}'

# 2. Synthesis Compile
curl -X POST http://localhost:8000/agent \
  -d '{"user_query":"Run synth_compile for design=\"des\" and return the log path."}'

# 3. Floorplan
curl -X POST http://localhost:8000/agent \
  -d '{"user_query":"Run floorplan for design=\"des\", top_module=\"des3\" and return the syn_ver and the path to floorplan.enc.dat."}'

# 4. Power Planning
curl -X POST http://localhost:8000/agent \
  -d '{"user_query":"Run powerplan for design=\"des\", top_module=\"des3\", impl_ver=\"cpV1_clkP1_drcV1__g0_p0\" and return the path to powerplan.enc.dat."}'

# 5. Placement
curl -X POST http://localhost:8000/agent \
  -d '{"user_query":"Run placement for design=\"des\", top_module=\"des3\", impl_ver=\"cpV1_clkP1_drcV1__g0_p0\" and return the path to placement.enc.dat."}'

# 6. Clock Tree Synthesis
curl -X POST http://localhost:8000/agent \
  -d '{"user_query":"Run CTS for design=\"des\", top_module=\"des3\", impl_ver=\"cpV1_clkP1_drcV1__g0_p0\" and return the path to cts.enc.dat."}'

# 7. Routing
curl -X POST http://localhost:8000/agent \
  -d '{"user_query":"Run routing for design=\"des\", top_module=\"des3\", impl_ver=\"cpV1_clkP1_drcV1__g0_p0\" and return the path to route_opt.enc.dat."}'

# 8. Save Design
curl -X POST http://localhost:8000/agent \
  -d '{"user_query":"Run save for design=\"des\", top_module=\"des3\", impl_ver=\"cpV1_clkP1_drcV1__g0_p0\" and return the archive path."}'

Project Structure

mcp-eda-example/
├── designs/                    # Design source files and results
│   ├── des/                   # DES encryption design
│   │   ├── rtl/              # RTL source files
│   │   ├── FreePDK45/        # Technology-specific results
│   │   │   ├── synthesis/    # Synthesis outputs
│   │   │   └── implementation/ # Physical implementation
│   │   └── config.tcl        # Design configuration
│   ├── b14/                  # VHDL benchmark design
│   ├── leon2/                # LEON2 processor design
│   └── des3/                 # Additional designs
├── server/                    # MCP server implementations
│   ├── synth_setup_server.py    # Synthesis setup service
│   ├── synth_compile_server.py  # Synthesis compile service
│   ├── floorplan_server.py      # Floorplanning service
│   ├── powerplan_server.py      # Power planning service
│   ├── placement_server.py      # Placement service
│   ├── cts_server.py           # Clock tree synthesis service
│   ├── route_server.py         # Routing service
│   ├── save_server.py          # Design save service
│   └── mcp/                   # MCP server implementations
│       ├── mcp_eda_server.py   # Main MCP server
│       └── start_mcp_server.sh # MCP server startup script
├── experiment/                 # Experimental framework
│   ├── generate/              # TCL generation methods
│   │   ├── baseline1_generate.py  # Pure LLM generation
│   │   ├── baseline2_generate.py  # LLM + template generation
│   │   ├── ours_generate.py       # MCP agent generation
│   │   └── common_config.py       # Shared configuration
│   ├── evaluate/              # Evaluation framework
│   │   ├── tcl_evaluator.py       # Main evaluator
│   │   └── evaluation_metrics.py  # Quality metrics
│   ├── results/               # Generation results
│   ├── evaluation_results/    # Evaluation reports
│   └── run_experiment.py      # Main experiment runner
├── scripts/                   # EDA tool scripts
│   ├── FreePDK45/           # Technology-specific scripts
│   │   ├── frontend/        # Synthesis scripts
│   │   └── backend/         # Physical implementation scripts
│   └── _helper/             # Python helper scripts
├── config/                    # Configuration files
│   ├── synthesis.csv           # Synthesis parameters
│   ├── imp_global.csv          # Implementation parameters
│   ├── placement.csv           # Placement parameters
│   └── cts.csv                 # CTS parameters
├── libraries/                 # Technology libraries
│   └── FreePDK45/           # FreePDK45 library files
├── logs/                      # Execution logs
├── deliverables/              # Final design outputs
├── mcp_agent_client.py          # Main MCP agent client
├── restart_servers.sh           # Server startup script
├── docker-compose.yml           # Docker Compose configuration
├── docker/                      # Docker configuration files
│   ├── Dockerfile.agent         # MCP Agent Client Dockerfile
│   ├── Dockerfile.servers       # MCP Servers Dockerfile
│   ├── health_check.py          # Health check script
│   ├── deploy.sh                # Deployment script
│   └── env.example              # Environment template
├── requirements.txt             # Python dependencies
└── README.md                    # This file

Configuration

Design Configuration (`designs/<design>/config.tcl`)

# Top-level module name
set TOP_NAME "your_module"

# File format (verilog/vhdl)
set FILE_FORMAT "verilog"

# Clock port name
set CLOCK_NAME "clk"

# Clock period in nanoseconds
set clk_period 1.0

Synthesis Configuration (`config/synthesis.csv`)

Parameter	Description	Default Value
`clk_period`	Target clock period (ns)	1.0
`DRC_max_fanout`	Maximum fanout	10
`DRC_max_transition`	Maximum transition time (ns)	0.5
`DRC_max_capacitance`	Maximum capacitance (pF)	5.0
`compile_cmd`	Compilation command	`compile_ultra`
`map_effort`	Mapping effort	`high`
`power_effort`	Power optimization effort	`medium`
`area_effort`	Area optimization effort	`medium`

Implementation Configuration (`config/imp_global.csv`)

Parameter	Description	Default Value
`ASPECT_RATIO`	Die aspect ratio	1.0
`target_util`	Target utilization	0.7
`design_flow_effort`	Flow effort level	`standard`
`design_power_effort`	Power optimization effort	`medium`

Experimental Framework

Supported Designs

The system supports various benchmark designs:

des: DES encryption design (Verilog)
b14: VHDL benchmark design
leon2: LEON2 processor design
des3: Additional DES variant

Experiment Methods

Baseline1 (Pure LLM): Direct GPT-4 TCL generation without templates
Baseline2 (LLM + Template): GPT-4 generation with template guidance
Ours (MCP Agent): Real MCP agent execution with EDA tools

Quality Metrics

The evaluation framework assesses TCL quality across multiple dimensions:

Syntax: Valid TCL syntax and structure
Completeness: Coverage of required commands
Executability: Ability to run successfully
Professionalism: Industry-standard practices

Running Experiments

# Complete experiment
cd experiment
python run_experiment.py --full

# Specific methods only
python run_experiment.py --generate baseline1 ours

# Evaluation only
python run_experiment.py --evaluate --summary

# Clean and run
python run_experiment.py --clean --full

API Reference

MCP Agent Client API

POST `/agent`

Main endpoint for natural language EDA operations.

Request Body:

{
  "user_query": "Run synth_setup for design=\"des\" and return the log path."
}

Response:

{
  "tool_called": "synth_setup",
  "tool_input": {
    "design": "des",
    "tech": "FreePDK45",
    "version_idx": 0,
    "force": true
  },
  "tool_output": {
    "status": "ok",
    "log_path": "/path/to/log/file",
    "reports": {
      "check_design.rpt": "report content"
    }
  }
}

Individual MCP Server APIs

Synthesis Setup Server (Port 13333)

POST /setup/run

{
  "design": "des",
  "tech": "FreePDK45",
  "version_idx": 0,
  "force": true
}

Floorplan Server (Port 13335)

POST /floorplan/run

{
  "design": "des",
  "top_module": "des3",
  "tech": "FreePDK45",
  "syn_ver": "cpV1_clkP1_drcV1",
  "g_idx": 0,
  "p_idx": 0,
  "force": true
}

Placement Server (Port 13337)

POST /place/run

{
  "design": "des",
  "top_module": "des3",
  "tech": "FreePDK45",
  "impl_ver": "cpV1_clkP1_drcV1__g0_p0",
  "g_idx": 0,
  "p_idx": 0,
  "restore_enc": "/path/to/powerplan.enc.dat",
  "force": true
}

Monitoring and Debugging

Log Files

All operations generate detailed logs in the logs/ directory:

logs/
├── setup/           # Synthesis setup logs
├── compile/         # Synthesis compile logs
├── floorplan/       # Floorplanning logs
├── powerplan/       # Power planning logs
├── placement/       # Placement logs
├── cts/            # Clock tree synthesis logs
├── route/          # Routing logs
└── save/           # Design save logs

Status Checking

# Check if all servers are running
netstat -tlnp | grep -E "(1333[3-9]|13440|8000)"

# Check server processes
ps aux | grep -E "(synth_setup|floorplan|placement|cts|route)"

# Monitor logs in real-time
tail -f logs/setup/des_setup_*.log

Common Issues and Solutions

1. Port Already in Use

# Kill existing processes
pkill -f "synth_setup_server.py"
pkill -f "floorplan_server.py"
# ... repeat for other servers

2. EDA Tool Not Found

# Check tool installation
which dc_shell
which innovus

# Set up environment variables
export SNPSLMD_LICENSE_FILE=/path/to/synopsys/license
export CDS_LIC_FILE=/path/to/cadence/license

3. Library Path Issues

# Verify FreePDK45 library
ls -la libraries/FreePDK45/

# Check library references in scripts
grep -r "FreePDK45" scripts/

Testing

Run Example Design

# Complete flow test
./run_pipeline.sh

# Individual stage test
python3 scripts/_helper/synthesis_config_row_no_dbg.py \
  --design des --version-idx 0 --tech FreePDK45

Test Different Designs

# Test VHDL design (b14)
curl -X POST http://localhost:8000/agent \
  -d '{"user_query":"Run synth_setup for design=\"b14\" and return the log path."}'

# Test another design (leon2)
curl -X POST http://localhost:8000/agent \
  -d '{"user_query":"Run synth_setup for design=\"leon2\" and return the log path."}'

Run Experiments

# Test experimental framework
cd experiment
python run_experiment.py --full --summary

# Test specific method
python run_experiment.py --generate ours --case_ids case_0

Contributing

We welcome contributions! Please see our for details.

Development Setup

Fork the repository
Create a feature branch
Make your changes
Add tests
Submit a pull request

Code Style

Use Python 3.9+ features
Follow PEP 8 style guidelines
Add type hints
Include docstrings for all functions

Documentation

: Step-by-step setup and usage
: Detailed API reference
: MCP protocol details

Acknowledgments

Synopsys for Design Compiler
Cadence for Innovus
OpenAI for GPT-4 API
FreePDK45 for open-source technology library
FastAPI for the web framework

Support

Issues: GitHub Issues
Discussions: GitHub Discussions
Documentation: Wiki

Version History

v2.0.0 - Added experimental framework and MCP server improvements
v1.2.0 - Enhanced monitoring and logging capabilities
v1.1.0 - Added VHDL support and improved error handling
v1.0.0 - Initial release with basic MCP EDA functionality

Made with for the EDA community

Author: Yiyi Lu, Jingyu Pan, Junyao Zhang

Email:

dukeceicenter/EDA-MCP-Server

Synopsys Design Compiler

Cadence Innovus

MCP EDA - Model Context Protocol for Electronic Design Automation

Features

System Architecture

MCP Server Architecture

Experimental Framework

Experiment Methods

Experiment Structure

Running Experiments

Prerequisites

Software Requirements

System Requirements

Installation

Option 1: Docker Deployment (Recommended)

Option 2: Local Installation

1. Clone the Repository

2. Set Up Python Environment

3. Configure Environment Variables

4. Verify EDA Tools Installation

Quick Start

1. Start MCP Agent Client

2. Start MCP Servers

3. Run Your First Design

Usage Guide

Natural Language Interface

Complete Design Flow Example

Project Structure

Configuration

Design Configuration (designs/<design>/config.tcl)

Synthesis Configuration (config/synthesis.csv)

Implementation Configuration (config/imp_global.csv)

Experimental Framework

Supported Designs

Experiment Methods

Quality Metrics

Running Experiments

API Reference

MCP Agent Client API

POST /agent

Individual MCP Server APIs

Synthesis Setup Server (Port 13333)

Floorplan Server (Port 13335)

Placement Server (Port 13337)

Monitoring and Debugging

Log Files

Status Checking

Common Issues and Solutions

1. Port Already in Use

2. EDA Tool Not Found

3. Library Path Issues

Testing

Run Example Design

Test Different Designs

Run Experiments

Contributing

Development Setup

Code Style

Documentation

Acknowledgments

Support

Version History

Design Configuration (`designs/<design>/config.tcl`)

Synthesis Configuration (`config/synthesis.csv`)

Implementation Configuration (`config/imp_global.csv`)

POST `/agent`