What else is being considered
The MCP server is one item on a longer list. The rest are AI-assisted features aimed at the part of this site people already complain about — a trace that comes back close but not clean. None of these are built, none are promised, and the ticks on the form are what decide which ones get made.
- MCP serverThe converters on this site, handed to your agent as tools it can call: trace a raster image to SVG, rasterise an SVG at a chosen width, reframe a landscape video to vertical. The work runs here, on the same pipeline the web pages use.
- AI SVG repairClean up a traced or hand-made SVG: merge paths that duplicate each other, smooth the stair-stepping a trace leaves on a curve, drop the stray specks a photo produces, and snap near-identical colours into one. The single most common thing people do by hand after a conversion.
- Describe an editSay what you want changed in plain language — "drop the background", "make the lettering one colour", "separate the badge from the text" — and get the SVG back edited. The layer panel already does the mechanical half of this by hand.
- Smart trace settingsChoose the preset and the colour count by looking at the image, instead of asking you to guess, squint at the result and run it again. The FAQ on every converter page currently exists to answer this question.
- Background removalCut the background before tracing rather than after. A white background traces to opaque white paths that a Cricut user has to find and delete, which is the first thing anyone making a cut file runs into.
- Alt text and layer namesGenerate alt text for the image and meaningful ids for the paths, so the SVG arrives readable by a screen reader and navigable in Figma instead of being forty shapes called path.
- Brand palette snappingMap the traced colours onto a palette you supply, so the output uses your four brand colours rather than the fourteen the JPEG happened to contain.
- Subject-aware reframingFollow the subject when reframing 16:9 to 9:16 instead of trusting a centre crop, so the person talking stays in frame when they move.
- Auto captionsBurn accurate captions into the vertical cut, since that is where most of these clips are going and they are watched with the sound off.
What it would do
MCP — the Model Context Protocol — is how an AI assistant discovers and calls tools that live outside itself. A PixelTools MCP server would expose the converters this site already runs: image-to-SVG tracing, SVG rasterising, and 16:9-to-9:16 video reframing. Your agent would call them the way it calls any other tool, get an artifact back, and carry on. The conversion itself would run here, on the same pipeline the web pages use — the same code, the same limits, the same hour-long artifact expiry.
Why it is not built
Because building it is the easy half. An MCP server over converters that already exist is a small piece of work; running it responsibly is not. Tracing an image and transcoding a video both cost real CPU, and agent traffic arrives in bursts rather than at human pace, so opening a server without per-caller limits and a way to identify callers would mean one enthusiastic script could take the converters down for everyone. That means accounts, or keys, or both — and that is a bigger decision than a weekend. The waitlist is how we find out whether it is worth making.
What the API does today
The converter endpoints this site runs on are currently reachable only from the PixelTools web app itself: requests from anywhere else are refused. That is deliberate while there is no way to tell one caller from another. An MCP server, or any documented public API, means reopening those endpoints on purpose, with limits attached — which is the same decision described above, seen from the other side.
Questions
Is the MCP server available now?
No. Nothing to install and no endpoint to point a client at. This page exists to gauge whether enough people want one to justify building it properly.
When will it launch?
There is no date, and we would rather say that than invent one. Joining the waitlist is what moves it up the list — if the list stays short, the honest answer is that it will not get built.
What will it cost?
Undecided. The converters on this site are free and stay free. An agent-facing server has a running cost that scales with use, so it may need limits or a paid tier above some threshold — but nobody on the waitlist will be billed for anything without being told first.
Which tools would it expose?
Most likely image-to-SVG first, since that is what people use this site for, followed by SVG-to-PNG and the vertical video reframer. The box on this page asking what you would point it at is what decides that order.
Can I use the HTTP API instead?
Not at the moment. The converter endpoints only accept calls from the PixelTools web app, because without a way to identify callers there is no way to keep one script from exhausting the queue. A documented API and the MCP server are the same decision and would arrive together.
What happens to my email address?
It is stored so we can tell you when the server opens, and used for nothing else — no newsletter, no sharing, no onward marketing. Ask and it is deleted.
Every PixelTools converter
Image
- Image → SVGAny raster format to vectors
- PNG → SVGTrace PNG into editable paths
- JPG → SVGTrace JPG into editable paths
- WebP → SVGTrace WebP into editable paths
- AVIF → SVGTrace AVIF into editable paths
- GIF → SVGTrace GIF into editable paths
- TIFF → SVGTrace TIFF into editable paths
- SVG → PNGRender vectors at any pixel size