MCP OpenVision

MCP OpenVision is an MCP server that provides image analysis capabilities powered by OpenRouter vision models. It connects client interfaces such as Claude Desktop and Cursor to OpenRouter's vision API, allowing users to send images for detailed visual inspection, optical character recognition, and scene interpretation. Developers, designers, and researchers use MCP OpenVision to delegate visual tasks directly to multimodal language models without manually opening external tools. The server enables clients to submit image inputs via base64 strings, web URLs, or local file paths, combined with specific text prompts and system instructions. It supports any vision-capable model available on OpenRouter, with defaults like Qwen 2.5 VL 32B Instruct and options to switch to Anthropic Claude 3.5 Sonnet or OpenAI GPT-4o. Through customizable temperature, token limits, and prompt roles, the server lets users extract tabular data from charts, inspect user interface mockups, read document scans, and perform specialized visual assessments directly within their chat environments.

Category: AI & LLM Tooling

Tags: image-analysis, multimodal, openrouter, vision

Visit MCP OpenVision

How to install and configure MCP OpenVision

1. Ensure you have an OpenRouter API key available. 2. Install MCP OpenVision via pip using uv pip install mcp-openvision or standard pip install mcp-openvision. 3. Alternatively, install automatically for Claude Desktop using Smithery with the command npx -y @smithery/cli install @Nazruden/mcp-openvision --client claude. 4. To configure manually for Claude Desktop or Cursor, open your client's configuration file (e.g., claude_desktop_config.json or .cursor/mcp.json). 5. Add the server under mcpServers with the command uvx, arguments ["mcp-openvision"], and provide your OPENROUTER_API_KEY and optional OPENROUTER_DEFAULT_MODEL in the env object. 6. Restart your MCP client to start using the image_analysis tool.

What you can do with MCP OpenVision

• Extract numerical data and trend summaries directly from screenshots of bar charts, line graphs, and financial tables. - Transcribe text, item names, and pricing from photographs of menus, receipts, or printed documents using vision models. - Audit user interface mockups and packaging designs to detect layout flaws, contrast issues, and visual hierarchy problems. - Identify objects, road signs, or equipment from local image files and explain their functions for educational workflows.

Key facts

• https://github.com/Nazruden/mcp-openvision
• AI & LLM Tooling, Design, Media & Creative
• image-analysis, multimodal, openrouter, vision

Part of MCP Servers

How do I install MCP OpenVision?

You can install MCP OpenVision using uv with uv pip install mcp-openvision or standard pip install mcp-openvision. For Claude Desktop, you can also install it automatically through Smithery by running npx -y @smithery/cli install @Nazruden/mcp-openvision --client claude.

What can MCP OpenVision do?

MCP OpenVision exposes an image_analysis tool that sends images to OpenRouter vision models. It accepts images via URLs, local file paths, or base64 strings, and lets you extract text, interpret charts, analyze designs, and inspect objects using custom prompts and system instructions.

Which vision models does MCP OpenVision support?

The server supports any vision-capable model accessible on OpenRouter. Its default model is qwen/qwen2.5-vl-32b-instruct:free, but you can also configure it to use models such as anthropic/claude-3-5-sonnet, anthropic/claude-3-opus, or openai/gpt-4o.

Which MCP clients work with MCP OpenVision?

MCP OpenVision works with any MCP-compatible environment that supports local servers, including Claude Desktop, Cursor, and the official Model Context Protocol Inspector tool for testing.

Is MCP OpenVision open source?

Yes, MCP OpenVision is open source and released under the MIT License. The code is publicly maintained on GitHub.