Florence-2 MCP Server

Name: Florence-2 MCP Server
Author: jkawamoto

@jkawamoto

代码仓库

关于 Florence-2 MCP Server

An MCP server for processing images using Florence-2

#python #florence-2 #mcp-server

基本信息

分类

其他

许可证

MIT

运行时

python

传输方式

stdio

发布者

jkawamoto

配置

使用下面的配置,将此服务器添加到你的 MCP 客户端。

{
  "mcpServers": {
    "florence-2": {
      "command": "uvx",
      "args": [
        "--from",
        "git+https://github.com/jkawamoto/mcp-florence2",
        "mcp-florence2"
      ]
    }
  }
}

工具

未检测到工具

工具是从 README 中自动提取的。维护者可以在 ## Tools 标题下列出工具,即可填充这部分内容。

概览

What is Florence-2 MCP Server?

An MCP (Model Context Protocol) server that processes images and PDF files using Microsoft’s Florence-2 model. It provides OCR to extract text and generates descriptive captions for images. Designed for developers and AI assistants (Claude, Goose, LM Studio) that need vision-based language model capabilities.

How to use Florence-2 MCP Server?

Install via the pre-built .mcpb bundle (for Claude Desktop) or by adding manual configuration entries to the claude_desktop_config.json, Goose config.yaml, or LM Studio MCP settings. All configurations use uvx to run the server from the Git repository. The server exposes two tools: ocr (extract text) and caption (generate captions), both accepting a src argument (file path or URL).

Key features of Florence-2 MCP Server

OCR for text extraction from images and PDFs
Image caption generation
Supports local file paths and remote URLs as input
Easy one‑click installs for Claude, Goose, and LM Studio
Runs via uvx for reproducible Python environments

Use cases of Florence-2 MCP Server

Extracting printed or handwritten text from scanned documents and photos
Generating alt‑text or summaries for images in a document workflow
Automating captioning of product images in an e‑commerce pipeline
Enabling voice‑assistant image analysis (e.g., reading a whiteboard photo aloud)
Integrating vision capabilities into LLM‑based AI agents without cloud dependencies

FAQ from Florence-2 MCP Server

What model does the server use?

It uses Microsoft’s Florence‑2 (Florence‑2‑large) hosted on Hugging Face.

How do I install it?

For Claude Desktop, download the .mcpb bundle from the Releases page and open it. For Goose, use the provided deeplink or edit ~/.config/goose/config.yaml. For LM Studio, click the “Add MCP Server” button in the README.

Can it process PDF files?

Yes. The OCR tool can extract text from PDFs as well as images.

What are the runtime requirements?

The server runs via uvx, so you need uv (or a Python environment with uvx). No additional Python package installation is required – uvx automatically fetches the dependencies.

Where does data processing happen?

All processing is performed locally on the user’s machine. The server reads image/PDF files from the local filesystem or fetches them from a provided URL.