Zovo Image MCP Server

Image generation and editing MCP for AI agents: create, resize, convert images locally. Local-first, free tier, $19 one-time Pro.

View on zovo.one

Documentation

All servers · Buy Pro $19 · All 46 for $39 · Source

Two ways to run it, both free to start. Open mcp.zovo.one/mcp/connect, copy the image URL and paste it into any client that takes a URL. It already carries a free token, so there is nothing to install, no account and no header to set. The URL is https://mcp.zovo.one/mcp/image/t/<token>. The token is not optional: the bare https://mcp.zovo.one/mcp/image connects and lists its tools and then answers every tool call with HTTP 401, so use the link from /mcp/connect or send Authorization: Bearer <token>. Or download image.mcpb and double-click it in Claude Desktop.

mcp-image

Say "make these five photos 1200 pixels wide" or "shrink this screenshot and strip the GPS out of it" and it happens, on your machine, in a second. This MCP server does the small image jobs that otherwise send you to a web uploader: resize, convert between PNG, JPEG, BMP, GIF and TIFF, compress with a real before-and-after byte count, crop, thumbnail a folder, watermark with your business name, and drop the EXIF block a phone camera writes into every photo. No upload, no account, no native dependency, no image editor.

In the official MCP Registry (io.github.theluckystrike/image-resize-convert-compress-watermark).

Listed on the AI Product Index — live remote endpoint at mcp.zovo.one/s/image, free tier, no signup.

image demo

The image chores of a freelance business, done from chat instead of from a browser tab you would rather not hand a client photo to.

60-second install

npm publish for @theluckystrike/mcp-image is pending. Until then, the .mcpb one-click bundle or a clone+build is the working path, both are verified below.

One-click (.mcpb): download image.mcpb from the latest release and double-click it in Claude Desktop: https://github.com/theluckystrike/mcp-servers/releases/latest

(claude_desktop_config.json):

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

Claude Code:

claude mcp add image -- npx -y @theluckystrike/mcp-image

(.cursor/mcp.json):

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

The npx form above starts working the moment the package is published. Until then, use the.mcpb bundle above, or build from source with exactly these three commands:

git clone https://github.com/theluckystrike/mcp-servers.git && cd mcp-servers
npm install
npm run build -w packages/mcp-license -w servers/image

Then point your client's command at node with one arg: the absolute path to servers/image/dist/index.js.

To run in Pro mode set MCP_LICENSE_KEY in the same config block, or call license_activate once with your key.

Tools

Tool	What it does
`image_info`	Format, pixel dimensions, megapixels, aspect ratio, file size in bytes, whether the image carries an alpha channel, and the bytes it spends per pixel
`image_resize`	A resized copy. `inside` fits within the box and keeps the ratio, `cover` fills the box and crops the overflow, `exact` stretches. Give one of width or height and the other follows
`image_convert`	Re-encode as PNG, JPEG, BMP, GIF or TIFF. Transparency going to JPEG is flattened onto white, and the answer says so
`image_compress`	Smaller bytes, with the count before and after and the percentage saved. `quality` (default 80), `max_width`, and `max_bytes` to hit a size target in one call
`image_crop`	Cut a rectangle: `x`, `y` from the top-left, `width`, `height`. A rectangle past an edge is refused with the real size, never clamped
`image_thumbnails`	One thumbnail per file into `out_dir`, each fitting inside `size` by `size` and keeping its aspect ratio
`image_watermark`	Text over the image at a corner and an opacity, white on a translucent plate. With no text, your business name from the shared profile mcp-invoice and mcp-docx write
`image_strip_metadata`	A copy with pixels and nothing else: EXIF, GPS, camera, capture time, XMP and colour profiles are not carried across
`image_batch_resize`	Resize a whole list into `out_dir`, each keeping its own aspect ratio, named `<name>-<W>x<H>.<ext>`
`image_dominant_colors`	The colours that cover most of an image, as hex codes with the share of pixels each covers. Free: 3; Pro: up to 16
`license_status`	Show free or Pro mode
`license_activate`	Activate a Pro key (verified offline)

Resource: image://recent returns the last 25 operations, newest first, with the inputs, the files written and when. Prompt: prepare_for_web takes a camera-sized image down to a page-sized JPEG, drops the metadata that came with it, and reports the bytes saved.

What you can say

You say	Tool
"How big is this image, and does it have transparency?"	`image_info`
"Make this 1200 pixels wide."	`image_resize`
"Turn this PNG into a JPEG."	`image_convert`
"This photo is 8 MB, get it under a megabyte."	`image_compress`
"Crop the top banner out of this screenshot."	`image_crop`
"Make thumbnails of everything in this folder."	`image_thumbnails`
"Put my company name in the corner of this photo."	`image_watermark`
"Strip the GPS coordinates before I send this."	`image_strip_metadata`
"Resize all of these to 1600 wide."	`image_batch_resize`
"What colours is this logo built from?"	`image_dominant_colors`
"Get this ready for the website."	`prepare_for_web` prompt

Worked example

You: This photo is too big for the site. Get it to 1600 wide and drop the metadata.

  image_info { path: "~/photos/DSC_0491.jpg" }
  -> 6000x4000, 24 MP, JPEG, 8.4 MB, no alpha

  image_compress {
    path: "~/photos/DSC_0491.jpg",
    max_width: 1600,
    quality: 80,
    max_bytes: 256000,
    out_path: "~/photos/DSC_0491-web.jpg"
  }
  -> 8.4 MB -> 243.7 KB (8809124 -> 249556 bytes, -97.2%)
  -> Method: JPEG quality 62 (searched down from 80 in 5 encodes to fit 250.0 KB)
     and a resize to 1600 px wide
  -> ~/photos/DSC_0491-web.jpg, 1600x1067, JPEG

The JPEG re-encode also drops the EXIF block the camera wrote, including the GPS coordinates. The original file is byte-for-byte unchanged.

Free vs Pro

	Free	Pro
`image_info`	Unlimited, at any size	Unlimited
`image_resize`, `image_convert`, `image_compress`, `image_crop`	Any source, writing outputs up to 4 MP (2000x2000 and a bit)	Any size, up to the 50 MB / 10,000 px input caps
`image_strip_metadata`	Any size	Any size
`image_thumbnails`, `image_batch_resize`	Up to 5 files per call, outputs up to 4 MP each	Any number of files, any size
`image_watermark`	Your business name from the shared profile	Any text you pass
`image_dominant_colors`	The top 3 colours	Up to 16

The free limit is on what gets written, not on what gets read: a 12 MP phone photo taken down to 1600 px wide is a free call, because that is the job. Writing 12 MP back out again is the Pro one.

A tier limit is an answer, not an error, and nothing is written when one refuses a call.

Pro is a one-time $19, or $39 for every server in the collection, lifetime.

quality is a JPEG parameter, and only a JPEG parameter

This is the one thing about image compression that surprises people, so the server says it in its own answer rather than accepting a number and quietly doing nothing with it. A JPEG throws away detail to get smaller, and quality is the dial for how much. A PNG is lossless: there is no dial. Passing quality: 40 to a PNG output does not make a PNG 60% smaller, it makes exactly the same PNG.

So the output format follows the extension of your out_path, and that is the lever:

out_path: "shot-small.jpg", quality applies, and a photo typically drops 80-95%.
out_path: "shot-small.png", quality is reported as not applicable, and only max_width removes bytes.

Palette quantisation was measured as an alternative for PNG and rejected: re-encoding a 300x220 noisy PNG through a 16-colour quantiser produced a larger file (115,451 bytes against 39,262), because the encoder here writes RGBA either way and quantising only destroys the row-to-row similarity that the deflate step was exploiting. A tool that claims to compress and returns a bigger file is worse than a tool that says which knob exists.

image_compress therefore always reports the byte count before and after, the percentage, and the method that did the work, and, if the output came out larger, it says so and tells you to keep the original.

Decompression bombs are refused from the header, before anything decodes

A 4 KB PNG can declare that it is 20,000 by 20,000 pixels. Decoding it allocates 1.6 GB of RGBA and takes the process down before any size check written after the decode could ever run. So the declared dimensions are read out of the container header, PNG IHDR, the JPEG SOF segment, the GIF logical screen descriptor, the BMP info header, the first TIFF IFD, and a file over 10,000 px on a side is refused there, with the memory it would have taken named in the message:

Error: /path/bomb.png declares 20000x20000 pixels in its PNG header and was refused before decoding:
this server caps a side at 10000 px. A small file that declares an enormous canvas is a decompression
bomb - decoding it would allocate 1526 MB of RGBA. Nothing was decoded.

The decoded size is checked again afterwards, because a header probe can come back empty on an unusual TIFF, and a guard that only sometimes runs is not a guard.

Existing files are never overwritten, and inputs are never modified

Every tool writes a new file and leaves its inputs byte-for-byte alone, there is no in-place mode, on purpose. An out_path that already exists is refused:

Error: /path/shot-small.png already exists and nothing was written.
Pass overwrite: true to replace it, or give a different out_path.

The path is reserved with an exclusive create, not an existence check, so two processes writing the same out_path at the same time cannot clobber each other: one wins, the other is refused and writes nothing. image_thumbnails and image_batch_resize reserve every one of their output paths before writing any of them, so a collision on file 3 does not leave files 1 and 2 behind as a half-done batch, and the reservations are released, so nothing empty is left on disk either.

Pass overwrite: true when replacing the file is what you want.

overwrite: true still does not let an output be an input. Writing a result back over one of its own sources destroys that source, the pixels are already decoded in memory and get written over the file they came from, so a 4000 px original becomes the 512 px thumbnail and every later read of that path is quietly wrong. So an out_path that resolves to (or shares an inode with) any input of the same call is refused before any work happens.

What "strip metadata" actually does

image_strip_metadata decodes the image and re-encodes it from the raw pixels. EXIF, GPS coordinates, the camera and lens, the capture time, XMP packets and embedded colour profiles are not removed one by one, they are simply never handed to the encoder, so they cannot come out the other side. Two consequences worth knowing:

The pixels are the same, the bytes are not. A JPEG goes through the encoder a second time, so the copy is not bit-identical to the original even though it looks the same. Keep the original if that matters.
Every other writing tool here has the same effect as a side effect. Resizing a photo also drops its EXIF.

Limits

Inputs over 50 MB are refused: an image decodes to width x height x 4 bytes of raw RGBA regardless of how well it is compressed on disk.
10,000 px per side, checked from the header first and from the decode second.
Read and written: PNG, JPEG, BMP, GIF and TIFF, detected by magic bytes rather than by file extension. No WebP, no AVIF, no HEIC, no SVG, there is no pure-JavaScript decoder for those that is worth shipping, and this server takes no native dependency.
An animated GIF is read as its first frame. This server does not do animation.
Watermark text is drawn with the bundled Open Sans bitmap faces (8, 16, 32, 64 and 128 px), and the largest one that fits the image is chosen. A character those faces do not carry is not drawn, so there is no CJK watermark.
No rotation, no filters, no colour correction, no OCR, no background removal. This server does the size-and-format jobs; it is not an image editor.

How it stores data

Only a register of what it did: ${XDG_DATA_HOME:-~/.local/share}/mcp-servers/image/operations.json, the last 500 operations, as plain JSON. Your images stay where you put them. Every write to the register runs inside an advisory lock on .../image/.lock, so two clients on one data directory cannot lose a record, and saves go to a temporary file and are renamed into place.

If the register is unreadable or not valid JSON it is never treated as "empty": it is moved aside byte-for-byte as operations.json.corrupt-<timestamp> with a marker beside it. The image you asked for is still written, the file is on disk before the register is touched, and the answer tells you the history could not be updated.

The watermark reads the one shared profile the whole suite uses, ${XDG_DATA_HOME:-~/.local/share}/mcp-servers/profile/business.json, written by business_set in mcp-invoice or mcp-docx. Nothing is invented: with no name stored and no text passed, image_watermark refuses and says which tool to run.

Privacy

All data stays local. The server reads and writes files on your machine and makes no network request of any kind -- not for licensing (keys are verified offline), not for fonts, not for telemetry. A client photo you resize is never uploaded anywhere, which is the entire reason this exists.

Pairs with

mcp-docx, size a logo here, then put it in the letterhead: image_resize to the header height, then the .docx writer picks the file up from disk.
mcp-pdf, shrink the scans before you merge them, so the joined PDF is not 40 MB.
mcp-resume, one correctly sized, metadata-free headshot for the CV and the application form.
office-suite, several servers behind one install, one config entry.

Troubleshooting

npx hangs or fails to find the package: npm publish for this package is pending. Use the .mcpb bundle or the clone-and-build path above until it lands.
"quality does not apply to a PNG output": write to a .jpg out_path, or pass max_width. See the section above.
"declares NxN pixels... and was refused before decoding": the file claims a canvas over 10,000 px per side. If it is genuinely that large, resize it with a tool that streams tiles; this server decodes whole images.
"does not start with the magic bytes": the file is not a PNG, JPEG, BMP, GIF or TIFF. A .png name on a WebP file is the usual cause; convert it first.
"already exists and nothing was written": pass overwrite: true, or a different out_path.
The watermark is tiny: the largest bundled face is 128 px, and it is chosen to fit within 80% of the image width. On a very large photo, resize first and watermark second.
Node version: requires Node >= 18. Check with node -v.

MIT licensed. Support: support@zovo.one

Built by theluckystrike.

First five minutes

Three prompts that scored 3 of 3, measured in round 11, 2026-09-04. Paste one in as it is written.

Here is a photo of mine pasted as base64. Save it as "photo", then make a web version 120 px wide and under 3 KB.

Paste this into Claude with the server connected.

What it did, measured in round 11, 2026-09-04: image_upload accepted the pasted base64 and named it photo.jpg (JPEG, 2574 bytes, format from magic bytes).

Here is my logo as base64. Save it as "logo", then convert it to JPEG on a white background.

Paste this into Claude with the server connected.

What it did, measured in round 11, 2026-09-04: One convert; the tool flattened the alpha onto white by itself and said so.

Here is my avatar as base64. Save it as "avatar", then crop it to a centered square and make 32 and 48 px thumbnails.

Paste this into Claude with the server connected.

What it did, measured in round 11, 2026-09-04: image_info showed 96x96, so the model said the centered crop is a no-op rather than pretending to crop, and produced both sizes as two separate download links.

On the free tier for this path: Info, resize, convert, compress, crop, strip metadata on images up to 4 MP; batches of 5.

Or the 46-server bundle for $39.

Set it up in your client

Exact config path, entry and caveats: Claude Desktop · Claude Code · Cursor · VS Code · Windsurf · Cline · Claude.ai and Claude Desktop connectors · all clients

Compared with the alternatives

MCP Image Tools vs Pictomancer and Image Resize API - which MCP server to pick · all comparisons

Guides

How to track billable hours inside Claude Code and Cursor · Create an invoice PDF from a chat message with an MCP server · Ask questions about an Excel or CSV file from Cursor or Claude · Watch a product price with Claude and get told when it drops · What the free tier includes and what Pro adds · Log expenses and mileage in Claude, split VAT, rebill to an invoice · Convert currencies in Claude with real ECB rates, no API key · Generate Word proposals and contracts from a chat message · Find a meeting time across time zones without doing the arithmetic · Write a resume and a cover letter from chat, without inventing anything · Bill a retainer on a schedule without a billing SaaS · Assemble a contract from your own clause library, in chat · Connect MCP servers to Claude.ai, Claude Desktop, Cursor and VS Code without installing anything · Merge, split and stamp PDFs from chat, and why some come back as glyph numbers · Read a.ics calendar in Claude: free and busy, conflicts, and billable meetings · Run a kanban board in Claude, with time tracking on the same task · Resize, compress and watermark images from a chat message · Categorize and reconcile a bank CSV export from chat · Send a quote from chat, then turn the yes into an invoice · Put a SEPA payment QR code on an invoice from chat · Zip and unzip archives safely from Claude or Cursor · Client deposits and retainers from chat, applied to your real invoices · Fixed assets and depreciation from chat, on the rates the tax authorities publish · Per diem and travel allowances from chat, on the rate tables the tax authorities publish · One double-entry ledger out of every server you already run · Loan and lease schedules from chat, closing exactly on zero · Work orders and job cards from chat, and why the markup goes on the unit cost · Price lists and rate cards from chat, and the 100x scale gap between the invoice and the quote · Change orders and the running contract value from chat, and why a changed line is two items · A petty cash float from chat, and why the cheque is not the sum of the vouchers · Client statements and payment chasers from chat, aged as at any date you name · Credit notes and purchase orders from chat, against your real invoices · One install, every server: the office-suite bundle · Close a month in chat: invoice, credit note, retainer, bank reconciliation, statement · MCP server not showing up in Claude Desktop: the six checks that find it · Where is claude_desktop_config.json, and what goes in it · claude mcp add: every flag, and the scope that silently loses your server · MCP servers on Windows: spawn npx ENOENT, backslashes and the PATH · Installing these MCP servers when npx does not work yet · Cursor MCP setup: mcp.json, the required type field, and where the servers appear · Which MCP servers work with no network at all · Invoicing an EU client with reverse charge, from a chat message · Quote, deposit, invoice, statement: the whole cycle in one conversation · Is this project making money? Hours against costs, without a spreadsheet · Building a tax year pack for your accountant from chat · Keeping a mileage log in chat, with the rate you actually claim · Reconciling a bank CSV against what you invoiced and spent · Rebilling client expenses, with a markup and the VAT handled honestly · Chasing an unpaid invoice: aging, then the letter · A bill of sale from chat, with the identifiers checked and the signature lines printed · Pricing a job from a rate card, and what to do when the scope changes · Working out your hourly rate from what you actually billed · Converting CSV to xlsx and back without opening Excel · Asking a spreadsheet questions in plain language · Merging a folder of receipts into one PDF from chat · Splitting one big scan into separate documents · Sending a month of paperwork as one archive, safely · MCP servers in VS Code: the one-word mistake that breaks every config · MCP servers in Windsurf and Cline: two defaults that waste an afternoon · Local MCP server or hosted URL: which one, and what you give up · What these MCP servers actually do on the free tier · Choosing an MCP server for invoicing: the seven questions worth asking · When you need an MCP server, and when a prompt is enough · MCP config file locations and JSON keys, every client · Why an MCP server does not appear, in order of likelihood · How MCP registry search actually works, measured · What is actually in the MCP registry: 6,000 rows counted · stdio or streamable HTTP: which MCP transport, and what breaks · MCP protocol versions: what each one changed · Every field in an MCP server config entry, by client · MCP config scopes: which definition wins when a server is defined twice · Where each MCP client writes its logs · Shipping an MCP server: a bundle to download or a URL to paste · What is inside a.mcpb MCP bundle · Charging for an MCP server: how licensing actually works · How much text an MCP server may return, and how much it may describe · What to check before letting an MCP server run · Why a directory says your hosted MCP server is not responding · The MCP registry binds one remote URL to one server name · Capital letters in your GitHub username change your MCP registry rank · How many MCP servers already have your word in the name · server.json field reference for the MCP registry · When an MCP tool fails, do not return a JSON-RPC error · structuredContent and outputSchema in MCP, and what MUST hold · Rules for naming an MCP tool, and the collision nobody plans for · MCP has no session, so how do two tool calls share state · x-mcp-header: mirroring tool parameters into HTTP headers · How a directory scores an MCP server, and why your worst tool decides it · What AI crawlers fetch that Googlebot does not, measured on 141 URLs · Search Console's URL Inspection API changes its answer between calls · Which MCP servers work by pasting a URL, with nothing installed · Where to find MCP servers that cost money, and how you actually pay · Getting an assistant to fill in a quote or estimate for a customer · Is there an MCP server for a petty cash book or a cash ledger · Currency conversion in an assistant, and which MCP server to use · Zipping and unzipping archives from an assistant, and what the guards do · Producing a delivery schedule or a work order document from a chat · Will an MCP server email the invoice to my client · Can an assistant read a photo of a receipt and log the expense · Which MCP server can generate invoice PDFs from a chat message? · Can an MCP server read a bank statement PDF and categorise the transactions? · Best MCP servers for small business accounting and paperwork in 2026 · MCP Supplier Directory: keep supplier lists from rotting in a spreadsheet · MCP Service Agreement: stop copying a rotting contract template off the internet · MCP Maintenance Log: know what service is due without anyone remembering · MCP Mileage Log: keep the deductible log at the moment of the drive, not in April · Invoice payment terms best practices: Net 30, due date and late fees · How to categorize bank transactions · Petty cash log template · All guides

More servers

MCP Per Diem Calculator — Employee Travel Allowances from Claude · MCP Server for Fixed Asset Register — Depreciation from Claude · MCP Server for Statements of Account — Customer Balances from Claude · MCP Server for Cash Book — Receipts, Payments and Balances from Claude · MCP Amortization · MCP Petty Cash Server — Track a Petty Cash Float from Claude · MCP Work Order · MCP Catalogue