STAMP

Overview

STAMP is an end-to-end, weakly-supervised deep learning pipeline designed for histopathology image analysis. It provides a complete workflow from whole-slide image processing to biomarker prediction without requiring pixel-level annotations. The tool supports multiple foundation models and can handle classification, regression, and survival analysis tasks. With MCP support, STAMP can be integrated into AI workflows as an intelligent tool for computational pathology research.

Try asking AI

After installing, here are 7 things you can ask your AI assistant:

• init

  Create a new STAMP configuration file

  Try:Initialize a new STAMP project configurationTry:Create a config file for my pathology study
• preprocess

  Preprocess whole-slide images into feature vectors

  Try:Preprocess histopathology slides for feature extractionTry:Convert WSI images to processed tiles
• encode_slides

  Encode patch-level features into slide-level embeddings

  Try:Extract slide-level features from patch embeddingsTry:Create slide representations using patch features
• encode_patients

  Encode features into patient-level embeddings

  Try:Aggregate slide features into patient-level representationsTry:Create patient embeddings from slide data
• train

  Train a Vision Transformer model

  Try:Train a deep learning model on my pathology dataTry:Build a biomarker prediction model from slides
• crossval

  Train a Vision Transformer model with cross validation

  Try:Perform cross-validation on my modelTry:Validate model performance with multiple folds
• deploy

  Deploy a trained Vision Transformer model

  Try:Deploy my trained model for predictionsTry:Make the model available for inference
• statistics

  Generate AUROCs and AUPRCs with 95%CI for a trained model

  Try:Calculate model performance metricsTry:Generate ROC curves for my biomarker predictor
• heatmaps

  Generate heatmaps for a trained model

  Try:Create heatmaps showing model attention regionsTry:Visualize which areas of the slide are most important

Note: Tool names extracted from CLI documentation but no MCP-specific tool signatures provided. MCP support is mentioned but not detailed in this README.

Comparable tools

Installation

# Install uv
curl -LsSf https://astral.sh/uv/install.sh | sh

# Clone and setup STAMP
git clone https://github.com/KatherLab/STAMP.git
cd STAMP

# GPU installation
uv sync --extra gpu
source .venv/bin/activate

# CPU installation
uv sync --extra cpu
source .venv/bin/activate

{
  "mcpServers": {
    "stamp": {
      "command": "uv",
      "args": ["run", "--package", "stamp", "--", "mcp"],
      "env": {
        "PYTHONPATH": "/path/to/STAMP"
      }
    }
  }
}

FAQ

What computational resources are required for STAMP?

STAMP can run on both CPU and GPU systems. GPU installation requires CUDA 13.0 and a compatible NVIDIA driver (version 580 or newer). The tool is designed to handle large datasets and can be deployed on HPC systems with SLURM.

How does STAMP's MCP integration work?

STAMP includes an 'mcp/' module that enables integration with Model Context Protocol. This allows the tool to function as an MCP server, exposing its computational pathology capabilities to AI workflows.

Compare STAMP with