TorchLeet MCP Server

An MCP (Model Context Protocol) server that turns any AI assistant into an interactive PyTorch interview coach using the TorchLeet problem set.

90 real PyTorch problems from ML/AI interviews at Google, Meta, Anthropic, and more — with progressive hints, learning paths, and company-specific prep plans.

Quick Start

Claude Desktop

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

{
  "mcpServers": {
    "torchleet": {
      "command": "npx",
      "args": ["-y", "torchleet-mcp"],
      "env": {
        "TORCHLEET_ROOT": "/path/to/your/TorchLeet/clone"
      }
    }
  }
}

Claude Code

claude mcp add torchleet -- npx -y torchleet-mcp

Set TORCHLEET_ROOT in your environment to enable notebook-based hints:

export TORCHLEET_ROOT=/path/to/your/TorchLeet/clone

Cursor / VS Code

Add to your MCP settings following your editor's MCP configuration format, using npx -y torchleet-mcp as the command.

What It Does

Tools

| Tool | Description | |------|-------------| | list_questions | Browse and filter questions by set, difficulty, company, category, or learning track | | get_question | Get full details for a question including Colab links | | get_learning_path | Structured learning paths: LLM path, basics, or advanced | | get_hint | Progressive hints (3 levels) — never reveals the full solution | | check_solution | Verification checklist for submitted code (does not execute) | | get_company_prep | Prioritized study plan for a specific company | | get_prerequisites | Find what to learn before tackling a hard problem |

Prompts

| Prompt | Description | |--------|-------------| | torchleet-tutor | Interactive study mode with Socratic questioning | | torchleet-interview-prep | Simulates a technical interview at a specific company | | torchleet-review | Code review focused on PyTorch best practices | | torchleet-explain | Deep-dive concept explanations with math and intuition |

Resources

| Resource | Description | |----------|-------------| | torchleet://questions | Full question catalog as JSON |

Environment Variables

| Variable | Required | Description | |----------|----------|-------------| | TORCHLEET_ROOT | No | Path to your TorchLeet repo clone. Enables notebook-based hints with problem statements, function signatures, and TODO guidance. Without it, the server still works using built-in descriptions. |

Example Usage

After adding the MCP server, try these in your AI assistant:

• "I'm preparing for an Anthropic interview, help me study"
• "Show me the LLM learning path"
• "Give me a hint for v3-14 (DPO Loss)"
• "List all expert-level questions"
• "What should I learn before tackling FlashAttention?"

Select the torchleet-tutor prompt for the best guided learning experience.

Development

git clone https://github.com/Exorust/TorchLeet.git
cd TorchLeet/mcp-server
npm install
npm run build

Test locally:

TORCHLEET_ROOT=.. node dist/src/index.js

License

MIT