mochify-cli CLAUDE.md

CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

Project

mochify-cli is a Rust CLI tool and MCP server that wraps the mochify.app image processing API (POST https://api.mochify.app/v1/squish). It uploads local images via multipart form and saves the processed result.

Commands

# Build
cargo build

# Build release
cargo build --release

# Check (fast, no binary output)
cargo check

# Run (process an image)
cargo run -- photo.jpg -t webp -w 800

# Process a PDF (auto-detected by .pdf extension)
cargo run -- document.pdf --op rasterize -t png --dpi 150
cargo run -- document.pdf -p "split into pngs"   # NLP prompt path

# Run MCP server on stdio
cargo run -- serve

# Run tests
cargo test

# Run a single test
cargo test <test_name>

# Lint
cargo clippy

# Format
cargo fmt

Architecture

src/
  main.rs        Async entry point. Parses CLI args (clap), dispatches:
                   - `serve` subcommand → starts MCP server on stdio
                   - no subcommand     → calls process_files(), which routes
                     `.pdf` inputs to process_pdfs() and images to the squish flow
  cli.rs         Clap `Args` struct and `Commands` enum (Serve subcommand)
  api.rs         `MochifyClient` + `ProcessParams` / `PdfParams` — all HTTP logic.
                   `squish()` posts an image to /v1/squish; `pdf()` posts a PDF to
                   /v1/pdf and saves the returned zip. Response bytes written to disk.
  mcp/
    mod.rs       Re-exports MochifyMcp
    tools.rs     `MochifyMcp` struct implements ServerHandler via rmcp macros.
                   Exposes `squish` (mirrors ProcessParams) and `pdf` (mirrors PdfParams).

Key design decisions

• Thin tools in MCP mode — the MCP client (e.g. Claude) handles natural-language interpretation and maps prompts to the structured squish / pdf tool parameters. No NLP layer needed in the CLI (the CLI's own --prompt flag does call /v1/prompt, including mode: "pdf" for PDFs).
• PDFs are auto-detected by the .pdf extension on the default path; PDFs and images can't be mixed in one invocation (the NLP prompt resolves to a single mode). --op split|rasterize (plus --dpi, -t, --quality for rasterize) configure it, or --prompt resolves it. Output is saved as a .zip.
• Auth is optional — without --api-key / MOCHIFY_API_KEY, requests go through on the free tier (25/month; unauthenticated IPs get 3/month). The key is sent as Authorization: Bearer <key>.
• rmcp macros pattern — tools use #[tool_router] on the impl block + #[tool_handler] on impl ServerHandler. The struct must have a tool_router: ToolRouter<Self> field initialized via Self::tool_router().