STAMP

STAMP requires OpenCV system libraries before Python packages can be installed. Clear the Triton cache if you have previously installed conflicting versions.

# Ubuntu 22.04+
sudo apt install libgl1 libglx-mesa0

# Clear Triton cache if needed
rm -rf ~/.triton

uv is the recommended dependency manager. Clone the STAMP repo and let uv handle the Python environment.

pip install uv
git clone https://github.com/KatherLab/STAMP.git
cd STAMP

Use uv to install the project in the current environment. Choose the CUDA variant for GPU support or cpu for CPU-only mode.

# GPU (CUDA)
uv pip install -e .[cuda]

# CPU only
uv pip install -e .[cpu]

Create a new configuration file for your analysis project. This generates a YAML config you can edit to point at your slide directory and labels.

stamp init --output-dir ./my_project

Run tile extraction and feature encoding on your whole-slide images. STAMP supports 20+ foundation models including Virchow-v2, UNI-v2, TITAN, and COBRA.

stamp preprocess --config ./my_project/config.yaml
stamp encode_slides --config ./my_project/config.yaml

After encoding, train a classification or survival model and produce explainability heatmaps for auditing.

stamp train --config ./my_project/config.yaml
stamp heatmaps --config ./my_project/config.yaml

Add STAMP's MCP server to your AI client configuration to enable conversational pipeline orchestration.

{
  "mcpServers": {
    "stamp": {
      "command": "uv",
      "args": ["run", "stamp", "mcp-server"],
      "cwd": "/path/to/STAMP"
    }
  }
}