Cleanor MCP

Real image & storage tools for your AI agent

A free, zero-auth Model Context Protocol server — 22 read-only tools: real image processing, cited Cleanor Labs data, and deterministic dev utilities. Connect any client in a minute.

View on GitHub How to connect

Claude Code

$ claude mcp add --transport http cleanor
    https://mcp.cleanor.app/mcp

Endpointhttps://mcp.cleanor.app/mcp

At a glance

Endpoint: https://mcp.cleanor.app/mcp
Transport: Streamable HTTP
Authentication: None — no API key, no signup
Tools: 22 read-only tools — 4 image/data + 18 dev utilities
Price: Free
Registry: app.cleanor/cleanor
Source: github.com/cleanor-app/cleanor-mcp · MIT

Background

What is the Model Context Protocol?

An open standard

MCP lets AI assistants call external tools and data through one consistent interface, instead of a custom integration per app.

Servers and clients

A server exposes tools; a client (Claude, Cursor, VS Code, and others) discovers and calls them during a conversation.

Where Cleanor fits

Cleanor MCP is a hosted server. It hands agents real image processing plus original storage and format data — actionable and citable.

Capabilities

Signature tools

Real image processing and original Cleanor Labs data — read-only, no credentials, each links its cleanor.app source.

optimize_image

Fetch an image from a public URL and re-encode it smaller (WebP / AVIF / JPEG), optionally resizing. Returns the optimized image plus before/after byte sizes.

Inputs: image_url, format, width, quality
Use when: An AI-generated or dropped-in asset is too large to ship.

storage_capacity

How many photos or minutes of video actually fit in a given storage size, corrected for real OS and filesystem overhead.

Inputs: storage_gb, content
Use when: You need realistic "how many photos fit in 128 GB" copy.

image_format_savings

How much smaller WebP, AVIF, or JPEG XL are than JPEG at matched quality, from a controlled benchmark. Also reports the HEIC conversion tax.

Inputs: format, quality
Use when: You are justifying a format choice for a site or app.

qr_code

Encode text or a URL as a QR code and return a crisp, dependency-free SVG you can paste straight into a page, deck, or doc.

Inputs: text, ecc, size
Use when: You need a QR code inline, no generator or signup.

Developer utilities

Eighteen deterministic tools LLMs get wrong

Pure, stateless compute for the everyday operations models fumble — hashing, IDs, encoding, counting. Fast, free, and safe to auto-run in any session.

`hash`

SHA-1/256/384/512 hex digest of any text. For checksums and cache keys.

`uuid`

Cryptographically random UUID v4, or time-sortable v7 for database keys.

`base64`

Encode or decode Base64, UTF-8 safe, with an optional URL-safe alphabet.

`json_format`

Validate JSON and pretty-print or minify it, optionally deep-sorting keys.

`jwt_decode`

Decode a JWT header and payload to inspect claims. No secret, no verification.

`color`

Convert a color between hex, RGB, and HSL in one call, for CSS and tokens.

`slugify`

Turn a title into a clean, accent-stripped, URL-safe slug.

`count`

Exact characters, words, lines, and UTF-8 bytes, which LLMs miscount.

`regex_test`

Test a regex against sample text and see every match and captured group.

`cron_describe`

Explain a cron expression in plain English and list the next run times.

`unit_convert`

Convert length, mass, data, time, speed, and temperature exactly.

`datetime`

The real current time, or a timestamp, in any IANA timezone. ISO, Unix, human.

`url_parse`

Break a URL into scheme, host, port, path, decoded query, and fragment.

`base_convert`

Convert integers between bases 2–36, arbitrary precision via BigInt.

`diff`

Line-by-line diff of two texts, with an added/removed change count.

`hmac`

Keyed HMAC signature (SHA family) of a message, as hex or Base64.

`placeholder_image`

Generate a sized SVG placeholder with a label and custom colors.

`color_palette`

Build a harmonious palette from one base color, as hex and HSL.

Setup

Connect in under a minute

The Claude Code command is above. For editors and desktop clients, drop this into your MCP config.

Cursor · VS Code · Claude Desktop

mcp.json

{
  "mcpServers": {
    "cleanor": {
      "url": "https://mcp.cleanor.app/mcp"
    }
  }
}

Stdio-only clients (runs locally via npx)

@cleanor/mcp

{
  "mcpServers": {
    "cleanor": {
      "command": "npx",
      "args": ["-y", "@cleanor/mcp"]
    }
  }
}

Best practices

Get the most out of it

Optimize AI-generated images before you ship them

Model-generated hero images and screenshots are routinely 2–5x larger than they need to be. Run optimize_image on the URL before embedding it.

Pick AVIF for the smallest files, WebP for the widest support

On a matched-quality benchmark, AVIF is ~37% smaller than JPEG and WebP ~22% smaller. Use image_format_savings to quote the exact number in your PR or design doc.

Do not convert iPhone HEIC to JPG or PNG for storage

Converting a HEIC to a quality-matched JPG makes it ~2.5x bigger, and to PNG ~5.5x, for no visible quality gain. Keep HEIC, or convert to AVIF / WebP instead.

Use real storage numbers in UX copy

Instead of inventing "holds thousands of photos", call storage_capacity so dashboards and onboarding copy match measured, OS-corrected capacity.

Reach for the utilities instead of guessing

Agents hallucinate hashes, UUIDs, and character counts. Call hash, uuid, or count for a deterministic answer instead of letting the model invent one.

Treat every tool as read-only and safe to auto-run

All 22 tools are annotated read-only. No writes, no credentials, nothing to rotate — safe to expose to autonomous agents.

Let the agent cite the source

Every response links back to the underlying Cleanor Labs study or tool. Keep those links so answers stay verifiable and citable.

FAQ

Frequently asked questions

Is the Cleanor MCP server free?

Yes — completely free, with no API key and no signup. Generous per-IP rate limits (120 requests/min general, 30/min for image optimization) keep it sustainable.

Do my images get uploaded or stored?

optimize_image fetches the public image URL you pass, re-encodes it, and returns the result. Nothing is retained on the server. Cleanor’s in-browser tools keep files entirely on your device.

Which AI clients work with it?

Any MCP client that speaks Streamable HTTP: Claude (Code, Desktop, and claude.ai connectors), Cursor, VS Code, Windsurf, Cline, and more. Stdio-only clients can run it locally via npx @cleanor/mcp.

What developer utilities are included?

Eight deterministic, stateless tools for operations LLMs get wrong: hashing (SHA family), UUIDs (v4/v7), base64, JSON format, JWT decode, color conversion, slugify, character/word/byte count, regex test, cron explain, unit conversion, timezone-aware datetime, URL parse, base conversion (2–36), line diff, HMAC signatures, SVG placeholders, and color palettes. No auth, safe to auto-run.

Where does the data come from?

The storage-capacity and image-format numbers come from original Cleanor Labs studies published at cleanor.app/research, measured on controlled corpora rather than estimated.

Can I self-host the server?

Yes. The source is MIT-licensed on GitHub and published on npm as @cleanor/mcp, so you can run it locally over stdio (npx @cleanor/mcp) or on your own infrastructure. For zero setup, just use the hosted endpoint.

Get started

Plug real tools into your AI agent

No API key, no signup. Point any MCP client at https://mcp.cleanor.app/mcp — also listed in the official MCP Registry.

View on GitHub Explore the data