App

Docs

MCP Integration

Call LearnRep as a native tool call — no shell required.

The @learnrep/mcp server exposes the same functionality as the CLI via the Model Context Protocol. If your harness supports MCP (Claude Code, Cursor, Windsurf), you can call generate_quiz as a tool call instead of running a shell command.

Run lr login first. The MCP server reuses the token stored at~/.learnrep/config.json.

Add to your MCP config

{
  "mcpServers": {
    "learnrep": {
      "command": "npx",
      "args": ["-y", "@learnrep/mcp"]
    }
  }
}

For Claude Code, add this to ~/.claude/settings.json under mcpServers.

For local development, set LEARNREP_API_URL=http://localhost:3456 in the MCP server environment.

Available tools

  • generate_quiz — generate a quiz from a topic or content string
  • share_quiz — make a quiz public and return the share URL
  • get_stats — return the user's streak, topics mastered, and avg improvement

generate_quiz parameters

{
  "topic": "react hooks",           // string, optional if content provided
  "content": "...",                 // string, optional — raw text to quiz on
  "focus": "useCallback vs useMemo",// string, optional
  "difficulty": "medium",           // easy | medium | hard | expert
  "count": 5,                       // number 1-20
  "types": ["multiple-choice"]      // array of question types
}

When to use MCP vs CLI

  • MCP — harness already has MCP support; cleaner tool-call interface; no shell dependency
  • CLI — simpler setup; works in any environment with a shell; easier to script

Both paths produce identical output. Use whichever fits your setup.