Skip to main content
Glama

Ribox MCP

Exact, file-based tools for AI assistants: the work a language model cannot do reliably by itself. Compress, resize and convert images, read or strip hidden metadata, make and read QR codes, count text exactly, diff text and validate JSON.

Files never leave your computer. Everything runs locally with WebAssembly codecs. No uploads, no network requests, no telemetry (the test suite checks this).

The same tools run in your browser at ribox.app.

Tools

Tool

What it does

compress_image

Make JPEG, PNG, WebP or AVIF files smaller (HEIC/BMP input accepted); target size such as "under 200 KB", max side length, format change

resize_image

Exact pixel sizes, fit/cover/stretch, crop, rotate 90/180/270, flip

convert_image

HEIC, JPEG, PNG, WebP, AVIF, BMP, ICO (favicon)

file_metadata

Read EXIF / GPS / XMP / IPTC / PNG text / PDF info, or remove it without re-encoding (verified afterwards)

generate_qr

Standards-compliant QR codes (link, text, Wi-Fi, vCard, email, SMS, WhatsApp, phone, location) as PNG or SVG

read_qr

Decode every QR code in photos or screenshots

count_text

Exact characters, words, sentences, lines, bytes; reading and speaking time

diff_text

Line, word or character diff with counts and a unified diff

check_json

Strict validation with the exact line and column of the first error; format, minify, sort keys

Not included on purpose: changing letter case, rewriting text and similar work an assistant already does well.

Related MCP server: Watermark & Resize Studio MCP Server

Install

Requires Node.js 22.12 or newer.

The npm package will be ribox-mcp. Until then, run it straight from this repository with npx github:AlgoWolfx/ribox-mcp. The first start downloads and builds it, which takes a minute; later starts are fast.

Claude Code

claude mcp add ribox -- npx -y github:AlgoWolfx/ribox-mcp

Claude Desktop

Open Settings > Developer > Edit Config and add this to claude_desktop_config.json, then restart Claude Desktop:

{
  "mcpServers": {
    "ribox": {
      "command": "npx",
      "args": ["-y", "github:AlgoWolfx/ribox-mcp", "/path/to/allowed/folder"]
    }
  }
}

On Windows, write the folder as "C:\\Users\\you\\Pictures".

Cursor

Add the same block to .cursor/mcp.json in your project (or ~/.cursor/mcp.json for all projects).

Other clients

Any MCP client that can start a stdio server works with the command npx -y github:AlgoWolfx/ribox-mcp [allowed-folder ...].

No local install? Use the remote endpoint

Assistants that cannot run local servers can connect to the remote endpoint https://ribox.app/api/mcp (Streamable HTTP). It offers only the four text tools (generate_qr, count_text, diff_text, check_json) and never accepts files. For anything that touches files, use the local server above.

Which files can it touch?

Only files inside the allowed folders:

  1. folders given as arguments (as in the examples above), otherwise

  2. the folders your client shares as MCP roots, otherwise

  3. the working directory the client starts the server in.

Paths outside these folders, including ../ tricks and symbolic links that point outside, are refused.

Results are written as new files next to the input (photo.min.jpg, photo.resized.png, photo.clean.jpg) or into output_dir. An original is replaced only when you pass overwrite: true. Up to 500 files per call; folders are processed non-recursively; each file up to 200 MB.

Privacy

  • Your files are read and written only on your machine.

  • The server makes no network requests and sends no telemetry. The test suite blocks fetch, http and https and fails if any of them is called.

  • The only usage signal is the public npm download count (once published).

Development

git clone https://github.com/AlgoWolfx/ribox-mcp.git
cd ribox-mcp
npm install        # also builds dist/
npm test           # core tests + in-memory MCP client tests
npm run build
node dist/index.js --help

Layout:

  • src/: the MCP server (stdio transport, file sandbox, tool handlers)

  • packages/core/: the environment-independent tool core shared with ribox.app; bundled into dist/index.js at build time

  • test/: end-to-end tests with an in-memory MCP client

Issues and pull requests are welcome.

License

MIT, see LICENSE. Third-party components and their licenses (including the LGPL-3.0 HEIC decoder, used as a separate, unmodified module) are listed in THIRD_PARTY_NOTICES.md.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      A
      quality
      D
      maintenance
      Provides AI agents with local file-processing capabilities for token counting, RAG chunking, CSV/JSON conversion, QR generation, and more, while keeping documents private on the user's machine.
      7
      1
      MIT
    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables AI assistants to search and run over 500 tools for PDF, text, data, crypto, network, and conversion tasks, with file processing done locally or at the edge so inputs are not uploaded.
      MIT
    • A
      license
      B
      quality
      B
      maintenance
      Enables AI assistants to inspect, convert, resize, crop, merge, extract and OCR local images, PDFs, audio and video entirely on the user's machine. Modules can be toggled individually, and all operations stay within a sandboxed working folder with atomic writes and no data sent anywhere.
      21
      65 PyPI
      MIT