Why CLI-first

The terminal is where agents already live.

LLM harnesses like Claude Code, Cursor, and Windsurf all have shell access. A CLI command is the lowest-friction integration possible — no SDK, no API key in the harness config, no extra auth flow. The agent just runs lr generate and gets back a URL.

The CLI is also composable. Pipe in any text, script it, or call it from a post-session hook. Because it reads from stdin, session transcripts, docs, and code files all work as input with zero preprocessing.

The web app and MCP server extend the CLI — they don't replace it. The MCP path is cleaner for harnesses that support tool calls natively, but the CLI works everywhere.