Wafer Map MCP Server

An MCP (Model Context Protocol) server that exposes semiconductor wafer analysis tools to AI assistants. These tools are designed for semiconductor test data analysis, enabling AI agents to correctly generate key engineering visualization charts such as wafer maps, P-charts, trend plots, and statistical analysis figures from raw test data.

Preview

Binary Map	Property Map	P-Chart

---

Tools

Tool	Description
run_wafer_analysis	Full analysis in one call: summary + binary map + all PIN maps + P-charts
get_wafer_info	Basic wafer summary (yield, pass/fail counts, PIN columns)
plot_wafer_bin	Binary pass/fail wafer map (BIN=0 → teal, else → black)
plot_wafer_property	Continuous-value heatmap for a single PIN column (blue → red)
plot_pchart	Normal probability plot per wafer for a PIN column

Data Format

CSV or ZIP (containing exactly one CSV) with columns:

BIN, X, Y, WAFER_ID, PIN_1, PIN_2, ..., PIN_N

• BIN = 0 → pass, otherwise fail
• X, Y → die coordinates on the wafer grid
• PIN_* → continuous measurement values

Colour Scale (IQR Robust Sigma)

Property maps and P-chart boundaries use IQR-based bounds to make subtle variations visible:

sigma  = (P75 - P25) / 1.35
IQR_L  = P50 - 6 × sigma
IQR_H  = P50 + 6 × sigma

---

Quick Start

Option A: Docker (recommended)

docker build -t wafer-mcp .
docker run -p 8001:8001 wafer-mcp

The server is now available at http://localhost:8001/mcp.

To analyze your own data files, mount a volume:

docker run -p 8001:8001 -v /absolute/path/to/data:/data wafer-mcp
# then pass file_path="/data/your_wafer.zip" when calling tools

Option B: Local Python

Requirements: Python 3.10+

pip install -r requirements.txt
python server.py

---

Sample Data

A sample dataset is bundled with the project at sample_data/sample_1.zip.

Location	Path
Local	./sample_data/sample_1.zip
Docker	/app/sample_data/sample_1.zip

Quick smoke test (Docker):

# inside the container the sample lives at /app/sample_data/sample_1.zip
# call any tool with this file_path to verify everything works

---

Claude Desktop Configuration

The server uses Streamable HTTP transport, so use the url form in claude_desktop_config.json:

{
  "mcpServers": {
    "wafer-map": {
      "url": "http://localhost:8001/mcp"
    }
  }
}

Steps:

1. Start the MCP server (local or Docker)
2. Add the config above to Claude Desktop
3. Restart Claude Desktop
4. Ask Claude to analyze a wafer data file — it will automatically pick the right tool

---

Tool Parameters

run_wafer_analysis

Param	Type	Default	Description
file_path	str	required	Path to .csv or .zip file
pin_columns	list[str] | None	None	Subset of PIN columns to plot; None = all
target_size	int	300	Output image pixel size

get_wafer_info

Param	Type	Default	Description
file_path	str	required	Path to .csv or .zip file

plot_wafer_bin

Param	Type	Default	Description
file_path	str	required	Path to .csv or .zip file
target_size	int	300	Output image pixel size

plot_wafer_property

Param	Type	Default	Description
file_path	str	required	Path to .csv or .zip file
pin_column	str	"PIN_1"	PIN column to visualise
target_size	int	450	Output image pixel size
data_l	float | None	None	Override lower bound of colour scale
data_h	float | None	None	Override upper bound of colour scale

plot_pchart

Param	Type	Default	Description
file_path	str	required	Path to .csv or .zip file
pin_column	str	"PIN_1"	PIN column to plot
target_size	int	300	Output image pixel size

---

Project Structure

.
├── server.py                    # MCP server entry point
├── requirements.txt             # Python dependencies
├── Dockerfile                   # Container definition
├── sample_data/
│   └── sample_1.zip             # Bundled sample wafer dataset
├── tools/
│   ├── workflow/
│   │   └── analyze_wafer.py     # Orchestrates full analysis
│   ├── information_read/
│   │   └── read_wafer_info.py   # Parse CSV/ZIP and compute yield
│   ├── wafer_map/
│   │   ├── wafer_bin_binary_plot.py    # Binary map renderer (PySide6)
│   │   └── wafer_item_property_plot.py # Property heatmap renderer (PySide6)
│   └── statistic_plot/
│       └── pchart_plot.py       # P-chart renderer (matplotlib)
└── pchart/
    └── PchartReportWidget.py    # Legacy Qt widget (reference only)

---

Tech Stack

• MCP: FastMCP — Streamable HTTP transport
• Wafer map rendering: PySide6 offscreen QPainter
• P-chart rendering: matplotlib + scipy + statsmodels

---

License

MIT