Modelscope Mcp Server

What is Modelscope Mcp Server?

A Model Context Protocol (MCP) server that integrates with the ModelScope ecosystem, providing seamless access to AI models, datasets, apps, papers, and generation capabilities through popular MCP clients.

How to use Modelscope Mcp Server?

Install and run via uvx modelscope-mcp-server or the pre-built Docker image spadrian/modelscope-mcp-server. Configure your MCP client (e.g., Claude Desktop, Cursor, VS Code, Cherry Studio) with the appropriate JSON, setting the environment variable MODELSCOPE_API_TOKEN to your ModelScope API token. The server supports stdio (default), HTTP, and SSE transports; for HTTP/SSE use --transport http and optionally --port.

Key features of Modelscope Mcp Server

• Retrieve information about the currently authenticated ModelScope user
• Generate images from text prompts or transform existing images (text-to-image and image-to-image)
• Search for machine learning models with advanced filtering (task type, author, inference support)
• Search for arXiv papers indexed in ModelScope with comprehensive metadata
• Documentation search and Gradio API integration (both coming soon)

Use cases of Modelscope Mcp Server

• Authenticate a user and retrieve their ModelScope account details
• Generate AI images from textual descriptions or modify existing images
• Discover and filter machine learning models hosted on ModelScope
• Find research papers on arXiv via the ModelScope index
• Integrate AI generation and search capabilities into MCP-compatible desktop clients

FAQ from Modelscope Mcp Server

What do I need to run Modelscope Mcp Server?

A valid ModelScope API token (set as MODELSCOPE_API_TOKEN), Python with uv or Docker to run the server, and an MCP-compatible client for interaction.

Which transports does Modelscope Mcp Server support?

It supports stdio (default), and HTTP/SSE transports. Use --transport http or --transport sse to switch, and optionally a custom port with --port.

Where does the data come from?

All data (models, papers, user info) is fetched from the ModelScope cloud platform. No local storage is used.

Is authentication required?

Yes, you must provide a ModelScope API token via the MODELSCOPE_API_TOKEN environment variable.

Are there any known limitations?

The README does not list any explicit limitations. Some features (documentation search, Gradio API integration) are marked as coming soon and are not yet available.