How to Use Mochify via MCP: AI Agent Image Compression (2026)

What is the Mochify MCP server?

MCP (Model Context Protocol) is a standard that lets AI assistants call external tools and services mid-conversation. We've built a Mochify MCP server that exposes our squish engine as a callable tool — meaning any AI assistant with MCP support can compress, resize, and optimize your images on your behalf.

This is the agent-first version of Mochify. The same compression quality and zero-retention privacy you get from the web tool at mochify.app, accessed directly from your AI workflow without opening a browser.

How the MCP connection works

When you describe an image optimization task to your MCP-connected assistant, it parses your prompt and identifies the compression intent. It then calls the Mochify MCP server with the appropriate flags — type, resize dimensions, quality, EXIF handling — and returns the optimized file back into your workflow.

The flow looks like this:

1. You describe the task in natural language to your AI assistant
2. The assistant identifies it as an image compression job
3. It calls the Mochify MCP tool with the right parameters
4. Our squish engine processes the image in memory
5. The optimized file is returned — no upload portal, no manual steps

The underlying engine is the same native C++ squish pipeline that powers the web tool. You're getting identical compression quality, just accessed from inside an AI agent context. (web.dev has a solid primer on why image optimization matters for Core Web Vitals if you want background on why this matters at scale.)

What the LLM controls: type, size, EXIF

The AI assistant connected via MCP makes the same decisions as the natural language feature on the web tool — but in an agentic, automated context. Here's exactly what it can set:

How to connect Mochify via MCP

The Mochify CLI includes the MCP server. Install it, then point your client at it.

1. Install the CLI

macOS (Homebrew):

brew tap getmochify/mochify
brew install mochify

Linux / WSL:

# x86_64 (most Linux / WSL)
curl -L https://github.com/getmochify/mochify-cli/releases/latest/download/mochify-linux-x86_64 -o mochify
chmod +x mochify && sudo mv mochify /usr/local/bin/

# arm64 (Raspberry Pi, AWS Graviton, etc.)
curl -L https://github.com/getmochify/mochify-cli/releases/latest/download/mochify-linux-arm64 -o mochify
chmod +x mochify && sudo mv mochify /usr/local/bin/

All binaries are statically linked — no dependencies needed. Also available on the GitHub releases page.

2. Add to your MCP client config

For Claude Desktop, edit ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "mochify": {
      "command": "mochify",
      "args": ["serve"]
    }
  }
}

To unlock your plan's limits, authenticate once with a browser login:

mochify auth login

This opens your browser, you sign in, and credentials are saved to ~/.config/mochify/credentials.toml. Both the CLI and the mochify serve MCP server pick them up automatically — no key to paste into your config. (MOCHIFY_API_KEY still works as an optional environment-variable override, but mochify auth login is the standard path.)

Restart Claude Desktop after saving. The Mochify squish tool will appear in your assistant's available functions — test it with: "Compress this image as a WebP for the web."

Hosted MCP: zero-install, any AI client

The Mochify MCP server is also available as a hosted HTTP endpoint at https://mcp.mochify.app. No binary to install, no Rust toolchain, no config file — paste the URL, authorize your account, and the squish tool appears in your AI assistant instantly.

One behavioural note specific to the hosted server: it returns the compressed image as a short-lived download URL on files.mochify.app (about a five-minute window), not as inline binary. Your original is still processed in RAM and discarded immediately; only the compressed output sits briefly in a pickup store so the link can resolve. The local CLI never uses that pickup store — it writes results straight to disk.

Claude.ai web (claude.ai)

In Claude.ai, go to Settings → Integrations and add a new integration with the URL:

https://mcp.mochify.app

Claude will redirect you to Mochify to sign in and authorize. After that, just tell Claude to compress an image by URL and it will call the squish tool automatically.

Claude Desktop (remote server)

Add a remote entry to your Claude Desktop config alongside any local servers you already have:

{
  "mcpServers": {
    "mochify-hosted": {
      "type": "http",
      "url": "https://mcp.mochify.app/mcp"
    }
  }
}

Claude Desktop will walk you through the OAuth flow the first time it connects. You can run both this remote entry and the local mochify serve entry at the same time — they use independent tokens.

Gemini, Cursor, Windsurf, and other MCP clients

Any client that supports remote MCP servers (HTTP+SSE or Streamable HTTP transport) works with the hosted endpoint. Add https://mcp.mochify.app/mcp as the server URL and follow the client's OAuth prompt.

Smithery & Glama

Mochify is listed on Smithery and Glama — MCP registries and marketplaces where you can discover and install MCP tools with a single click. Search for "Mochify" to add the hosted server directly from the registry without touching any config file.

Example: compressing images from a prompt

Say you're working in Claude Desktop and you say:

"Can you compress this product photo as an Instagram-ready WebP and strip the EXIF data?"

Behind the scenes, the assistant calls Mochify's MCP server with roughly:

• --type webp
• --resize 1080x1080
• --strip-exif

The result comes back in seconds — a web-optimized 1080×1080 WebP with no metadata attached. You didn't touch a settings panel or look up Instagram's recommended dimensions. The agent handled the full decision chain.

MCP vs. the web tool: when to use each

Both options use the same squish engine and deliver the same output quality. The difference is workflow fit.

If you're optimizing images as part of a larger agent workflow — content pipelines, automated publishing, code-adjacent tasks — MCP is the right path. For quick, individual jobs, the web tool is faster to reach.

Cheat sheet: prompt-to-flag mapping

FAQ

Related guides

• How to Self-Host an Image Optimization API with Docker — Full control, local deployment
• The 2026 Guide to Next-Gen Formats: WebP, AVIF, and JPEG XL — Format benchmarks and use cases
• Privacy & Image Optimization: A Comprehensive Guide (2026) — Zero-retention explained in depth
• The Risks of EXIF Data in Image Compression (2026) — Why metadata stripping matters