three.ws Vision runs on your own machine — the client starts it, so there is no endpoint to ping. 134 installs a week from npm. Last commit 1 Nov 2026.
Image understanding for AI agents — analyze and describe any image via the three.ws pipeline.
We read the source, 1 d ago · rules 995398eba69f
A value the model can set ends up inside a file or shell call. That is not a flaw by itself — for a terminal server it is the job — but it is where things go wrong when it is not.
await fs.writeFile(out, bytes);
Things with no honest explanation: a promise that contradicts the code, code that runs at install time while hiding what it does, data leaving the machine.
'http://metadata.google.internal/computeMetadata/v1/instance/service-accounts/default/token',
What this server is able to do. For an MCP server this is often the job itself — a terminal server runs commands because that is what it is for. Listed so you know what you are plugging in, not as an accusation.
const r = await redis.eval(
filePath = path.join(ROOT_PUBLIC, url);
const child = spawn(command[0], command.slice(1), { stdio: 'inherit', shell: false });
exec(compile(job["code"], "<three-ws-blender-mcp>", "exec"), namespace)
const [row] = await sql`SELECT voice_provider, voice_id FROM agent_identities WHERE id = ${agentId}`.catch(() => []);
return subprocess.run(
if (reset) { state.cursor = null; state.total = 0; grid.innerHTML = skeletons(); stateEl.hidden = true; }
'UklGRmQBAABXQVZFZm10IBAAAAABAAEAQB8AAIA+AAACABAAZGF0YUABAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA…
const TELEGRAM_API = 'https://api.telegram.org';
Is this your server and something here is wrong? Tell us — corrections are free and do not require a plan.
That is not a flaw by itself — but it is where things go wrong when it is not the job. We re-read this code on every release. Watch it and you hear from us the day another one appears.
This server runs on your own machine — install it with the package manager and the client starts it for you. Package name taken from the official registry entry.
claude mcp add vision-mcp -- npx -y @three-ws/vision-mcp
{
"mcpServers": {
"vision-mcp": {
"command": "npx",
"args": [
"-y",
"@three-ws/vision-mcp"
]
}
}
}
[mcp_servers.vision-mcp]
command = "npx"
args = ["-y", "@three-ws/vision-mcp"]
{
"mcpServers": {
"vision-mcp": {
"command": "npx",
"args": [
"-y",
"@three-ws/vision-mcp"
]
}
}
}
{
"servers": {
"vision-mcp": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"@three-ws/vision-mcp"
]
}
}
}
Optional settings from the registry entry, left out of the block above:
THREE_WS_BASE (Base URL of the three.ws API that serves the vision endpoint. Override only when self-hosting or targeting a preview deployment), THREE_WS_TIMEOUT_MS (Per-request timeout in milliseconds for the vision call (the VLM may chain across free lanes before answering)), THREE_WS_API_KEY (Optional three.ws API key (bearer token). Not required — the vision endpoint serves anonymous callers on a free per-IP rate limit. Set it to raise your rate limit to the per-user tier. Never needed to access the free model).
Video scene understanding for AI agents via the Primate Vision API.
Image & PDF tools for AI agents: compress, convert, resize, PDF, AI vision, pipeline.
The image pipeline for agents: presets, assets and jobs. Answers with asset ids and stable URLs.
Video pipeline for AI agents: analyze with AI, transcode to MP4/HLS/DASH, edit, protect and deliver.
Shrink images to a token budget for vision LLM agents and pipelines. Not for chat-pasted images.
UI design from prompts, screenshots, and URLs for AI coding agents and theme tokens.
Turn videos, PDFs, screenshots, and articles into anchored, cached understanding for agents.
MCP server for AI-powered image recognition and description using OpenAI vision models.
Answers built from our own checks of this server.