gifcompressors MCP Server

A Model Context Protocol server for AI agents, shipped as one self-contained file. It gives an agent four GIF tools that run on the user's own machine with the gifsicle engine — files are never uploaded anywhere. Same engine and same results as the in-browser tools on this site.

Tools

• compress_gif — mode quality (20–35% smaller) / balanced (35–55%) / maximum (50–70%), optional target_kb budget; writes name-compressed.gif next to the input
• resize_gif — by exact width or percentage, animation and timing preserved
• convert_gif — to MP4 or WebP (uses ffmpeg when present; otherwise points the user to the free in-browser converters on this site)
• get_recommendations — reads a file's size/frames/dimensions and returns per-platform settings (Discord, Twitter/X, email, emoji) plus the matching page URLs

Requirements

• Node.js ≥ 18 (no npm install, no dependencies — the file is self-contained)
• gifsicle: the server uses, in order, the GIFSICLE_PATH environment variable, a gifsicle on PATH, or the gifsicle npm package if the project already has it
• Files up to 100 MB per operation; everything is processed locally and in place

Install (two commands)

mkdir -p ~/.gifcompressors
curl -o ~/.gifcompressors/gifcompressors-mcp.mjs \
  https://www.gifcompressors.com/mcp/gifcompressors-mcp.mjs

Then register it with your MCP client (Claude Desktop, Claude Code, Cursor, Qoder, Codex, …) — in Claude-style config JSON:

{
  "mcpServers": {
    "gifcompressors": {
      "command": "node",
      "args": ["/ABSOLUTE/PATH/TO/.gifcompressors/gifcompressors-mcp.mjs"]
    }
  }
}

Verify it works

After connecting, ask the agent to call get_recommendations on any GIF. Or check the transport directly:

echo '{"jsonrpc":"2.0","id":1,"method":"tools/list"}' \
  | node ~/.gifcompressors/gifcompressors-mcp.mjs

Source & license

MIT-licensed, built from github.com/jiusongvip/gif-compressor/mcp-server. This page's file is regenerated from that source on every site deploy, so it always matches the latest release (v0.1.0).