Env-Doctor

The missing link between your GPU and Python AI libraries

"Why does my PyTorch crash with CUDA errors when I just installed it?"

Because your driver supports CUDA 11.8, but pip install torch gave you CUDA 12.4 wheels.

Env-Doctor diagnoses and fixes the #1 frustration in GPU computing: mismatched CUDA versions between your NVIDIA driver, system toolkit, cuDNN, and Python libraries.

It takes 5 seconds to find out if your environment is broken - and exactly how to fix it.

Doctor "Check" (Diagnosis)

Features

Installation

pip install env-doctor

Or with uv (a faster Python package manager):

# Install as an isolated tool (won't touch your project env)
uv tool install env-doctor

# Or run once without installing
uvx env-doctor check

Both methods install the same package from PyPI — pick whichever you prefer.

MCP Server (AI Assistant Integration)

Env-Doctor includes a built-in Model Context Protocol (MCP) server that exposes diagnostic tools to AI assistants like Claude Code and Claude Desktop.

Quick Setup for Claude Desktop

1. Install env-doctor:

   pip install env-doctor
   

2. Add to Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json):

   {
     "mcpServers": {
       "env-doctor": {
         "command": "env-doctor-mcp"
       }
     }
   }
   

3. Restart Claude Desktop - the tools will be available automatically.

Available Tools (11 Total)

• env_check - Full GPU/CUDA environment diagnostics
• env_check_component - Check specific component (driver, CUDA, cuDNN, etc.)
• python_compat_check - Check Python version compatibility with installed AI libraries
• cuda_info - Detailed CUDA toolkit information
• cudnn_info - Detailed cuDNN library information
• cuda_install - Step-by-step CUDA installation instructions
• install_command - Get safe pip install commands for AI libraries
• model_check - Analyze if AI models fit on your GPU
• model_list - List all available models in database
• dockerfile_validate - Validate Dockerfiles for GPU issues
• docker_compose_validate - Validate docker-compose.yml for GPU configuration

Demo — Claude Code using env-doctor MCP tools

Example Usage

Ask your AI assistant:

• "Check my GPU environment"
• "Is my Python version compatible with my installed AI libraries?"
• "How do I install CUDA Toolkit on Ubuntu?"
• "Get me the pip install command for PyTorch"
• "Can I run Llama 3 70B on my GPU?"
• "Validate this Dockerfile for GPU issues"
• "What CUDA version does my PyTorch require?"
• "Show me detailed CUDA toolkit information"

Learn more: MCP Integration Guide

Usage

Diagnose Your Environment

env-doctor check

Example output:

🩺 ENV-DOCTOR DIAGNOSIS
============================================================

🖥️  Environment: Native Linux

🎮 GPU Driver
   ✅ NVIDIA Driver: 535.146.02
   └─ Max CUDA: 12.2

🔧 CUDA Toolkit
   ✅ System CUDA: 12.1.1

📦 Python Libraries
   ✅ torch 2.1.0+cu121

✅ All checks passed!

On new-generation GPUs (e.g. RTX 5070 / Blackwell), env-doctor catches architecture mismatches and distinguishes between two failure modes:

Hard failure — torch.cuda.is_available() returns False:

🎯  COMPUTE CAPABILITY CHECK
    GPU: NVIDIA GeForce RTX 5070 (Compute 12.0, Blackwell, sm_120)
    PyTorch compiled for: sm_50, sm_60, sm_70, sm_80, sm_90, compute_90
    ❌ ARCHITECTURE MISMATCH: Your GPU needs sm_120 but PyTorch 2.5.1 doesn't include it.

    This is likely why torch.cuda.is_available() returns False even though
    your driver and CUDA toolkit are working correctly.

    FIX: Install PyTorch nightly with sm_120 support:
       pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu126

Soft failure — torch.cuda.is_available() returns True via NVIDIA's PTX JIT, but complex ops may silently degrade:

🎯  COMPUTE CAPABILITY CHECK
    GPU: NVIDIA GeForce RTX 5070 (Compute 12.0, Blackwell, sm_120)
    PyTorch compiled for: sm_50, sm_60, sm_70, sm_80, sm_90, compute_90
    ⚠️  ARCHITECTURE MISMATCH (Soft): Your GPU needs sm_120 but PyTorch 2.5.1 doesn't include it.

    torch.cuda.is_available() returned True via NVIDIA's driver-level PTX JIT,
    but you may experience degraded performance or failures with complex CUDA ops.

    FIX: Install a newer PyTorch with native sm_120 support for full compatibility:
       pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu126

Check Python Version Compatibility

env-doctor python-compat

🐍  PYTHON VERSION COMPATIBILITY CHECK
============================================================
Python Version: 3.13 (3.13.0)
Libraries Checked: 2

❌  2 compatibility issue(s) found:

    tensorflow:
      tensorflow supports Python <=3.12, but you have Python 3.13
      Note: TensorFlow 2.15+ requires Python 3.9-3.12. Python 3.13 not yet supported.

    torch:
      torch supports Python <=3.12, but you have Python 3.13
      Note: PyTorch 2.x supports Python 3.9-3.12. Python 3.13 support experimental.

⚠️   Dependency Cascades:
    tensorflow [high]: TensorFlow's Python ceiling propagates to keras and tensorboard
      Affected: keras, tensorboard, tensorflow-estimator
    torch [high]: PyTorch's Python version constraint affects all torch ecosystem packages
      Affected: torchvision, torchaudio, triton

💡  Consider using Python 3.12 or lower for full compatibility

💡  Cascade: tensorflow constraint also affects: keras, tensorboard, tensorflow-estimator

💡  Cascade: torch constraint also affects: torchvision, torchaudio, triton

============================================================

Get Safe Install Command

env-doctor install torch

⬇️ Run this command to install the SAFE version:
---------------------------------------------------
pip install torch torchvision --index-url https://download.pytorch.org/whl/cu118
---------------------------------------------------

Install CUDA Toolkit

Display instructions or execute the installation directly:

# Show platform-specific steps (default)
env-doctor cuda-install

# Preview what would run — no changes made
env-doctor cuda-install --dry-run

# Execute interactively (asks [y/N] before running)
env-doctor cuda-install --run

# Execute headlessly — great for CI/scripts
env-doctor cuda-install --run --yes

# Install a specific version, headless
env-doctor cuda-install 12.6 --run --yes

Example dry-run output (Windows):

[DRY RUN] [1/1] winget install Nvidia.CUDA --version 12.2

[DRY RUN] [1/1] nvcc --version

CUDA 12.2 installation completed successfully.
Verification: PASSED

Full log: C:\Users\you\.env-doctor\install.log

Every run writes a timestamped log to ~/.env-doctor/install.log for debugging.

Supported Platforms:

• Ubuntu 20.04, 22.04, 24.04
• Debian 11, 12
• RHEL 8, 9 / Rocky Linux / AlmaLinux
• Fedora 39+
• WSL2 (Ubuntu)
• Windows 10/11 (via winget)
• Conda (all platforms)

Exit codes for CI pipelines:

Install Compilation Packages (Extension Libraries)

For extension libraries like flash-attn, SageAttention, auto-gptq, apex, and xformers that require compilation from source, env-doctor provides special guidance to handle CUDA version mismatches:

env-doctor install flash-attn

Example output (with CUDA mismatch):

🩺  PRESCRIPTION FOR: flash-attn

⚠️   CUDA VERSION MISMATCH DETECTED
     System nvcc: 12.1.1
     PyTorch CUDA: 12.4.1

🔧  flash-attn requires EXACT CUDA version match for compilation.
    You have TWO options to fix this:

============================================================
📦  OPTION 1: Install PyTorch matching your nvcc (12.1)
============================================================

Trade-offs:
  ✅ No system changes needed
  ✅ Faster to implement
  ❌ Older PyTorch version (may lack new features)

Commands:
  # Uninstall current PyTorch
  pip uninstall torch torchvision torchaudio -y

  # Install PyTorch for CUDA 12.1
  pip install torch --index-url https://download.pytorch.org/whl/cu121

  # Install flash-attn
  pip install flash-attn --no-build-isolation

============================================================
⚙️   OPTION 2: Upgrade nvcc to match PyTorch (12.4)
============================================================

Trade-offs:
  ✅ Keep latest PyTorch
  ✅ Better long-term solution
  ❌ Requires system-level changes
  ❌ Verify driver supports CUDA 12.4

Steps:
  1. Check driver compatibility:
     env-doctor check

  2. Download CUDA Toolkit 12.4:
     https://developer.nvidia.com/cuda-12-4-0-download-archive

  3. Install CUDA Toolkit (follow NVIDIA's platform-specific guide)

  4. Verify installation:
     nvcc --version

  5. Install flash-attn:
     pip install flash-attn --no-build-isolation

============================================================

Check Model Compatibility

env-doctor model llama-3-8b

🤖  Checking: LLAMA-3-8B (8.0B params)

🖥️   Your Hardware: RTX 3090 (24GB)

💾  VRAM Requirements:
  ✅  FP16: 19.2GB - fits with 4.8GB free
  ✅  INT4:  4.8GB - fits with 19.2GB free

✅  This model WILL FIT on your GPU!

List all models: env-doctor model --list

Cloud GPU Recommendations:

# Get cloud GPU recommendations for a model that doesn't fit
env-doctor model llama-3-70b --recommend

# Direct VRAM lookup (no model name needed)
env-doctor model --vram 80000 --recommend

☁️   Cloud GPU Recommendations

  FP16 (~140.0 GB):
    $27.20 /hr  azure  ND96asr_v4              8x A100 (40GB each)          180.0GB free
    $29.39 /hr  gcp    a2-highgpu-8g            8x A100 (40GB each)          180.0GB free
    ...

Automatic HuggingFace Support (New ✨)
If a model isn't found locally, env-doctor automatically checks the HuggingFace Hub, fetches its parameter metadata, and caches it locally for future runs — no manual setup required.

# Fetches from HuggingFace on first run, cached afterward
env-doctor model bert-base-uncased
env-doctor model sentence-transformers/all-MiniLM-L6-v2

Output:

🤖  Checking: BERT-BASE-UNCASED
    (Fetched from HuggingFace API - cached for future use)
    Parameters: 0.11B
    HuggingFace: bert-base-uncased

🖥️   Your Hardware:
    RTX 3090 (24GB VRAM)

💾  VRAM Requirements & Compatibility
  ✅  FP16:  264 MB - Fits easily!

💡  Recommendations:
1. Use fp16 for best quality on your GPU

Validate Dockerfiles

env-doctor dockerfile

🐳  DOCKERFILE VALIDATION

❌  Line 1: CPU-only base image: python:3.10
    Fix: FROM nvidia/cuda:12.1.0-runtime-ubuntu22.04

❌  Line 8: PyTorch missing --index-url
    Fix: pip install torch --index-url https://download.pytorch.org/whl/cu121

More Commands

CI/CD Integration

Generate a GitHub Actions workflow with one command:

env-doctor init --github-actions

This creates .github/workflows/env-doctor.yml — review, commit, and push. Your CI will validate the ML environment on every push and PR.

Or add manually:

# JSON output for scripting
env-doctor check --json

# CI mode with exit codes (0=pass, 1=warn, 2=error)
env-doctor check --ci

GitHub Actions example:

- run: pip install env-doctor
- run: env-doctor check --ci

Documentation

Full documentation: https://mitulgarg.github.io/env-doctor/

• Getting Started
• Command Reference
• MCP Integration Guide
• WSL2 GPU Guide
• CI/CD Integration
• Architecture

Video Tutorial: Watch Demo on YouTube

Contributing

Contributions welcome! See CONTRIBUTING.md for details.

License

MIT License - see LICENSE