The same engine, server-side.

Anonymous requests work, with tight limits. A key lifts them and is sent as a bearer token. A key we do not recognise is refused with 401 rather than quietly downgraded to the anonymous tier — a silent downgrade turns a typo into a rate-limit error an hour later.

Authorization: Bearer $VECTORTRACE_KEY

Vectorize one raster image. Send it as a multipart file part or as base64 inside a JSON body. The response is the vector file itself, or the document as JSON when you ask for it.

PNG · JPG · JPEG · WebP · BMP · GIF

Request

Examples

curl -X POST "https://vectortrace.app/api/v1/vectorize?format=svg" \
  -H "Authorization: Bearer $VECTORTRACE_KEY" \
  -F "image=@logo.png" \
  -F 'options={"preset":"logo","colors":8}' \
  -o logo.svg

import { readFile, writeFile } from 'node:fs/promises'

const form = new FormData()
form.append('image', new Blob([await readFile('logo.png')]), 'logo.png')
form.append('options', JSON.stringify({ preset: 'logo', colors: 8 }))

const response = await fetch('https://vectortrace.app/api/v1/vectorize?format=svg', {
  method: 'POST',
  headers: { Authorization: `Bearer ${process.env.VECTORTRACE_KEY}` },
  body: form,
})
if (!response.ok) throw new Error(await response.text())
await writeFile('logo.svg', Buffer.from(await response.arrayBuffer()))

import json, os, requests

with open("logo.png", "rb") as image:
    response = requests.post(
        "https://vectortrace.app/api/v1/vectorize",
        params={"format": "svg"},
        headers={"Authorization": f"Bearer {os.environ['VECTORTRACE_KEY']}"},
        files={"image": image},
        data={"options": json.dumps({"preset": "logo", "colors": 8})},
    )
response.raise_for_status()
open("logo.svg", "wb").write(response.content)

JSON body

POST /api/v1/vectorize?format=svg
Content-Type: application/json
Accept: application/json

{
  "image": "iVBORw0KGgoAAAANSUhEUgAA…",
  "options": { "preset": "logo", "colors": 8 }
}

Every field is optional. A preset fills in the rest; a field you set wins over the preset. These are the same controls as the converter's inspector, and the same table the Rust engine reads.

Success

{
  "document": {
    "version": 1, "width": 512, "height": 512, "unit": "px",
    "palette": ["#101114", "#e8491d"],
    "elements": [{ "kind": "path", "fill": 0, "stroke": null, "subpaths": [ … ] }]
  },
  "stats": { "paths": 12, "nodes": 184, "colors": 2, "ms": 41, "estimatedSvgBytes": 3210 }
}

Every answer carries the rate-limit headers, refused or not. X-RateLimit-Limit, X-RateLimit-Remaining, X-RateLimit-Reset

Error envelope

One envelope, always. Branch on the code, never on the prose — the code is stable and the message is not.

{ "error": { "code": "IMAGE_TOO_LARGE", "message": "…" } }

An image over the pixel limit is refused, not downscaled. The converter in your browser reduces oversized artwork because the alternative is a quarter-gigabyte allocation in your own tab; here the ceiling is a budget, and returning coordinates that do not map onto the artwork you sent would be worse than an error.

The reference above is generated from the same zod schemas the server validates with, so there is no hand-written spec to fall out of date. Point a client generator at it.

GET https://vectortrace.app/api/openapi.json

/api/openapi.json

A plain-text summary of the product, the pages and this endpoint, written for language models and agents. The full version adds the page index and the preset table.

GET https://vectortrace.app/llms.txt
GET https://vectortrace.app/llms-full.txt

The engine as a command. Same presets, same output, no server in the loop.

npx @vectortrace/cli in.png -o out.svg --preset laser

An MCP server exposing the engine to agents over stdio or streamable HTTP. Three tools: