MCP Florence2
Run Microsoft Florence-2 locally to read, caption and detect objects in images and PDFs
- Pricing
- Open source
- Platforms
- CLI, macOS, Windows
- Runs
- Locally
- Version
- 0.5.4
- Listing updated
- today
Similar MCP servers
About MCP Florence2
MCP Florence2 is an open-source MCP server for ai models and llm tooling. It runs locally on your machine and is started by your AI client.
Florence-2 MCP Server gives an assistant local eyes without calling a vision API. It runs Microsoft's Florence-2 model on your own machine, so images and documents are not sent to any outside service and there are no per-token fees. Inputs can be local paths or http(s) URLs for common image formats or multi-page PDFs, processed page by page.
Three tools are exposed. ocr returns all visible text, caption writes a detailed description, and process runs a raw Florence-2 task prompt such as plain or region-aware OCR, object detection with bounding boxes, dense region captions or region proposals. It detects Apple Silicon (MPS), NVIDIA CUDA or falls back to CPU, and by default inference runs in a subprocess that frees VRAM afterwards; --cache-model keeps it loaded for speed. The --model flag can switch to the smaller Florence-2-base.
Install paths include an .mcpb bundle for Claude Desktop, a one-click link for LM Studio, a Goose extension, or uvx for any stdio client. The README warns that URL inputs can enable server-side request forgery, so untrusted environments should route outbound traffic through a filtering proxy.
Install
Run in a terminal
claude mcp add mcp-florence2 -- npx -y https://github.com/jkawamoto/mcp-florence2/releases/download/v0.5.4/mcp-florence2.mcpbAdd to claude_desktop_config.json
{
"mcpServers": {
"mcp-florence2": {
"command": "npx",
"args": [
"-y",
"https://github.com/jkawamoto/mcp-florence2/releases/download/v0.5.4/mcp-florence2.mcpb"
]
}
}
}Add to ~/.cursor/mcp.json
{
"mcpServers": {
"mcp-florence2": {
"command": "npx",
"args": [
"-y",
"https://github.com/jkawamoto/mcp-florence2/releases/download/v0.5.4/mcp-florence2.mcpb"
]
}
}
}Add to .vscode/mcp.json
{
"servers": {
"mcp-florence2": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"https://github.com/jkawamoto/mcp-florence2/releases/download/v0.5.4/mcp-florence2.mcpb"
]
}
}
}Add to ~/.codeium/windsurf/mcp_config.json
{
"mcpServers": {
"mcp-florence2": {
"command": "npx",
"args": [
"-y",
"https://github.com/jkawamoto/mcp-florence2/releases/download/v0.5.4/mcp-florence2.mcpb"
]
}
}
}Add to your client's MCP settings (most use this shape)
{
"mcpServers": {
"mcp-florence2": {
"command": "npx",
"args": [
"-y",
"https://github.com/jkawamoto/mcp-florence2/releases/download/v0.5.4/mcp-florence2.mcpb"
]
}
}
}Using several servers? Build one config file with the MCP config generator.
Tools this server provides (3)
Taken from the project documentation; the running server may differ.
- ocr
- Extract all visible text from an image or PDF
- caption
- Generate a detailed caption for an image
- process
- Run a custom Florence-2 task prompt such as object detection or region captioning
What it does
- Local OCR
- Extracts text from images and multi-page PDFs.
- Detailed captions
- Describes objects, layout, colours and background.
- Raw task prompts
- Run OCR with regions, object detection or region proposals.
- Hardware detection
- Uses Apple Silicon, NVIDIA CUDA or CPU.
- Memory release
- Frees VRAM after each call unless caching is on.
- One-click installs
- Bundle for Claude Desktop, link for LM Studio, Goose extension.
Good for
- Read text from a scanned invoice without uploading it
- Describe a photo for an assistant that lacks vision
- Find objects and bounding boxes in an image
- Extract text from a multi-page PDF
Pricing
Open source
Open source; runs locally with no API fees
MCP Florence2 FAQ
Does it send images to a cloud service?
No. The README says processing runs locally on your machine.
Can it read PDFs?
Yes, multi-page PDFs are handled page by page.
How do I use a lighter model?
Pass --model florence-community/Florence-2-base.
Is there a security caveat?
Yes. URL inputs can expose a server to SSRF, so untrusted setups should filter outgoing HTTP traffic.
Is MCP Florence2 free?
Yes. MCP Florence2 is open source, so you can use and self-host it for free.
How do I add MCP Florence2 to Claude Code?
Run this command in your terminal: claude mcp add mcp-florence2 -- npx -y https://github.com/jkawamoto/mcp-florence2/releases/download/v0.5.4/mcp-florence2.mcpb
What are the best alternatives to MCP Florence2?
The closest alternatives to MCP Florence2 are headroom, omniget, next-ai-draw-io, mcp-for-blender and Claude Code Ultimate Guide.
Where this comes from
- Description: written from the project's README.
- Install steps: from the official MCP Registry entry.



