openai-vision-mcp-server
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
openai-vision-mcp-server
GitHub Repository: https://github.com/ygq-future/openai-vision-mcp-server
A Model Context Protocol (MCP) stdio server for bounded, high-precision image analysis using OpenAI Chat Completions-compatible Vision APIs (e.g. OpenAI gpt-4o, Qwen VL, DeepSeek Vision, Local vLLM/Ollama, etc.).
✨ Features
Multi-Source Image Inputs: Analyze images directly from
file://local paths,http:///https://URLs, orbase64raw data payloads.Smart Adaptive Tiling & Overview Pipeline: Automatically generates overview thumbnails and ordered overlapping detail tiles for high-resolution images, preserving visual detail without hitting token limits.
Universal OpenAI Compatibility: Works with any API endpoint following the standard OpenAI
/chat/completionsvision protocol.Configurable Security & Resource Bounds:
Optional SSRF protection for public-network-only deployments (
VISION_ALLOW_PRIVATE_NETWORK=false).Optional local file root restrictions (
VISION_ALLOWED_FILE_ROOTS).Configurable ceilings for file size, decoded pixel count, HTTP timeouts, and max redirects.
Clean Stdio Transport: Keeps
stdoutstrictly isolated for standard MCP JSON-RPC protocol messages while outputting diagnostics tostderrwithout leaking credentials or payloads.
Local-file and private-network access are permissive by default for local MCP convenience. SetVISION_ALLOWED_FILE_ROOTS to explicit roots and VISION_ALLOW_PRIVATE_NETWORK to false when the MCP client or
analyzed prompts are not fully trusted.
Related MCP server: Vision MCP
🚀 Quick Start
You can run openai-vision-mcp-server without manual installation using npx or bunx.
The executable also resolves correctly when npm launches the current checkout through a .bin directory link, so an
explicit version such as @latest is optional.
MCP Client Integration Examples
Add the server to your preferred MCP client's configuration file (e.g., Claude Desktop, Cursor, Windsurf, VS Code / Antigravity):
Standard claude_desktop_config.json / mcp.json:
{
"mcpServers": {
"vision": {
"command": "npx",
"args": ["-y", "openai-vision-mcp-server"],
"env": {
"VISION_BASE_URL": "https://api.openai.com/v1",
"VISION_API_KEY": "your-api-key-here",
"VISION_MODEL": "gpt-4o"
}
}
}
}⚙️ Environment Variables & Configuration
Configuration is passed entirely through environment variables defined in the MCP server configuration:
Environment Variable | Type | Required | Default | Description |
| String | Yes | — | Base URL of the OpenAI-compatible API (e.g., |
| String | Yes | — | API key for authentication. |
| String | Yes | — | Vision model name exposed by the configured OpenAI-compatible endpoint (e.g., |
| Integer | No |
| Default hard ceiling for detail tiles (1 to 64). |
| String | No |
| Optional delimiter-separated path whitelist for |
| Boolean | No |
| Set to |
| Integer | No |
| Max raw image download size in bytes (default: 20MB). |
| Integer | No |
| Max allowed total decoded image pixels (default: 100MP). |
| Integer | No |
| HTTP request timeout in milliseconds (30s). |
| Integer | No |
| Maximum HTTP redirect count. |
| Integer | No |
| Max concurrent tile processing calls (default 1 to prevent 429 rate limits). |
🛠 Available Tools
analyze_images
Analyzes single or multiple images using configured Vision models and produces structured analysis reports. The MCP declaration marks the Tool as read-only and non-destructive, non-idempotent because each call can consume upstream API usage, and open-world because it can fetch remote images and call an external Vision endpoint.
Input Schema
Property | Type | Description |
|
| The query or instruction for the vision analysis. |
|
| List of image objects to analyze (1 to 10). |
|
| Tiling strategy ( |
|
| Override hard ceiling for detail tiles for this call (1 to 64). |
ImageSource Types
File Source:
{ "type": "file", "uri": "file:///path/to/image.png", "label": "optional label" }URL Source:
{ "type": "url", "url": "https://example.com/photo.jpg", "label": "optional label" }Base64 Source:
{ "type": "base64", "data": "<base64_string>", "mediaType": "image/png", "label": "optional label" }
Results, warnings, and errors
A successful result includes complete. When complete is false, the answer may still contain useful evidence, but
every warnings[] entry explains the missing coverage with these machine-readable fields:
{
"code": "TILE_BUDGET_EXCEEDED",
"message": "Detail coverage requires 16 tiles, but this call allows 10.",
"retryable": false,
"userActionRequired": true,
"nextAction": "Continue with the partial result, disclose the missing coverage, and increase maxTiles only if the user requests complete analysis.",
"details": { "requiredTiles": 16, "allowedTiles": 10 }
}A failed Tool call returns isError: true, the same guidance as readable text, and
structuredContent.error containing code, message, retryable, userActionRequired, nextAction, and optional
safe details. The calling AI should follow nextAction: permanent input/configuration/protocol failures explicitly
say not to retry, while transient network, timeout, rate-limit, and server failures allow one bounded caller retry. If
the same transient failure repeats, stop and notify the user instead of looping.
Errors never include credentials, authorization headers, Base64/image bytes, upstream response bodies, complete local paths or URLs, stack traces, or full prompts.
💻 Local Development
This project uses Bun for fast testing and compilation.
# Install dependencies
bun install
# Run unit and integration tests
bun test
# Run code check (format, lint, typecheck, test, build)
bun run check
# Build output files
bun run build📄 License
MIT License © 2026 ygq-future
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
This server cannot be deployed
Maintenance
Related MCP Connectors
OCR, transcription, file extraction, and image generation for AI agents via MCP.
Focused MCP server for OpenAI image/audio generation (v2.0.0). Wraps endpoints via HAPI CLI.
Generate images from Claude, ChatGPT, Cursor and Codex with 12 top AI image models. Up to 4 images per call, SVG vectorization, background removal, upscaling and image-to-video. Remote MCP with OAuth, no API keys.
Generate images from Claude, ChatGPT, Cursor and Codex with 12 top AI image models and 10 tools: up to 4 images per call, SVG vectorization, editing and more. Remote MCP with OAuth, no API keys. Pay per image, no subscription.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceMCP server for analyzing images using multiple vision LLM providers (OpenCode, OpenAI, Anthropic, Google, and custom OpenAI-compatible endpoints). Provides tools to analyze single or multiple images, list providers, and test vision capabilities.MIT
- AlicenseAqualityCmaintenanceEnables any MCP client to perform image understanding and OCR via any OpenAI-compatible vision-language model. Supports local, private inference without images leaving the machine.227 npmMIT
- AlicenseAqualityDmaintenanceProvides image recognition capabilities to MCP clients by integrating with OpenAI-compatible vision models, supporting local images, URLs, multi-image comparison, and model listing.4MIT
- AlicenseNot gradedqualityCmaintenanceEnables MCP clients to gain vision capabilities by analyzing images, extracting text, and comparing images through any OpenAI-compatible vision endpoint.24 npmMIT