three.ws Vision — Claude Code MCP server: how to install
Examines images, extracts screenshot text, and generates captions through visual models. Runs over stdio without billing or API credentials.
claude mcp add --env THREE_WS_BASE=<THREE_WS_BASE> --env THREE_WS_TIMEOUT_MS=<THREE_WS_TIMEOUT_MS> --env THREE_WS_API_KEY=<THREE_WS_API_KEY> --transport stdio vision-mcp -- npx -y @three-ws/vision-mcpWhat three.ws Vision does
This package gives assistants three inspection tools: analyze_image, describe_image, and get_vision_status. The first answers visual questions and extracts printed copy from screenshots. The second crafts alt text or descriptive summaries with selectable detail levels. The third checks backend availability before requests go out.
Incoming requests hit NVIDIA NIM vision endpoints first, followed by automated fallback handling. You supply graphic data as base64 payloads or public HTTPS links. File size cannot exceed 12 MB, with JPEG, PNG, WebP, and GIF accepted.
Private IP ranges and localhost URLs fail because the remote service downloads external links directly. You need Node.js 20 or newer on your machine. Basic requests share an IP rate limit, which you can upgrade by setting THREE_WS_API_KEY.
What it helps with
- Extract printed text from screenshots via OCR
- Write descriptive alt text for UI graphics
- Critique three-dimensional character renders
- Inspect visual content hosted on a public web address
Example prompts for the agent
- Read the text on the logo at
https://three.ws/logo.pngusinganalyze_image. - Generate a brief alt text description for this base64 image string.
- Call
get_vision_statusto see if the vision endpoint is ready for incoming requests.
How to install three.ws Vision
Platform CLI
claude mcp add --env THREE_WS_BASE=<THREE_WS_BASE> --env THREE_WS_TIMEOUT_MS=<THREE_WS_TIMEOUT_MS> --env THREE_WS_API_KEY=<THREE_WS_API_KEY> --transport stdio vision-mcp -- npx -y @three-ws/vision-mcpReplace <…> with your own values
Platform CLI
codex mcp add vision-mcp --env THREE_WS_BASE=<THREE_WS_BASE> --env THREE_WS_TIMEOUT_MS=<THREE_WS_TIMEOUT_MS> --env THREE_WS_API_KEY=<THREE_WS_API_KEY> -- npx -y @three-ws/vision-mcpReplace <…> with your own values
Config file
{
"mcpServers": {
"vision-mcp": {
"command": "npx",
"args": [
"-y",
"@three-ws/vision-mcp"
],
"env": {
"THREE_WS_BASE": "<THREE_WS_BASE>",
"THREE_WS_TIMEOUT_MS": "<THREE_WS_TIMEOUT_MS>",
"THREE_WS_API_KEY": "<THREE_WS_API_KEY>"
}
}
}
}Replace <…> with your own values
Config file
{
"servers": {
"vision-mcp": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"@three-ws/vision-mcp"
],
"env": {
"THREE_WS_BASE": "<THREE_WS_BASE>",
"THREE_WS_TIMEOUT_MS": "<THREE_WS_TIMEOUT_MS>",
"THREE_WS_API_KEY": "<THREE_WS_API_KEY>"
}
}
}
}Replace <…> with your own values
Platform CLI
gemini mcp add -e THREE_WS_BASE=<THREE_WS_BASE> -e THREE_WS_TIMEOUT_MS=<THREE_WS_TIMEOUT_MS> -e THREE_WS_API_KEY=<THREE_WS_API_KEY> vision-mcp npx -y @three-ws/vision-mcpReplace <…> with your own values
Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about three.ws Vision
How do I install three.ws Vision in Claude Code?
Run claude mcp add vision-mcp -- npx -y @three-ws/vision-mcp in your terminal. It registers the server over stdio without extra configuration.
What does three.ws Vision do?
It allows coding agents to inspect image files, perform visual OCR, and compose captions. NVIDIA NIM models process the media through the service backend.
Does three.ws Vision require an API key?
You do not need a key because requests are free under the per-IP quota. Adding THREE_WS_API_KEY is only necessary if you need higher per-user rate limits.
