local-llm-mcp-geotools

EmreCandan0/local-llm-mcp-geotools

3.1

If you are the rightful owner of local-llm-mcp-geotools and would like to certify it and/or have it hosted online, please leave a comment on the right or send an email to dayong@mcphub.com.

A local AI-powered geospatial analysis system using a locally hosted LLM and MCP server for processing natural language requests.

Tools
4
Resources
0
Prompts
0

local-llm-mcp-geotools

Locally running LLM (Ollama + Phi3) integrated with MCP server for geospatial analysis, TIFF cropping, NDVI & DEM computation.

A local AI-powered geospatial analysis system that uses a locally hosted LLM (Ollama + Phi3) and a Model Context Protocol (MCP) server to process natural language requests. The assistant understands user prompts and calls specialized tools for:

  • GeoTIFF metadata analysis
  • Image cropping by geographic coordinates
  • NDVI (Normalized Difference Vegetation Index) computation
  • DEM (Digital Elevation Model) value retrieval

🔧 Key Features

  • 🧠 Local LLM integration (Ollama + Phi3) — No internet or API keys required
  • 🔌 MCP server-based tool-calling architecture
  • 🗺️ Geospatial image processing: TIFF/JP2 analysis, NDVI, DEM, and cropping
  • 💬 Natural language interface — Users interact with simple text prompts
  • 📂 FastAPI + FastMCP backend with GDAL for geospatial operations

🖼️ Example Use Cases

  • “Analyze this TIFF file and show me the coordinates.”
  • “Crop the image between these lat/lon values.”
  • “Give me the NDVI value at this point.”
  • “What is the DEM elevation at this coordinate?”

🚀 Getting Started

1. Clone the Repository

git clone https://github.com/emrecandan0/local-llm-mcp-geotools.git

cd local-llm-mcp-geotools

  1. Start the MCP Server Make sure ports and paths are set correctly in the code. python mcp_server.py

  2. Launch the LLM Assistant (CLI) Make sure Ollama is running and the Phi3 model is pulled. ollama run phi3 python client.py

⚙️ Requirements Python 3.9+

Ollama (with phi3 model)

GDAL

FastAPI

NumPy

Requests

MCP Server (via fastmcp)

📁 Project Structure

├── client.py # Main CLI assistant (LLM + tool-calling)

├── mcp_server.py # FastAPI MCP tool server with GDAL tools

├── functions/

│ └── funcs_pool.py # Helper functions: geometry, EPSG, metadata

├── static/outputs/ # Output folder for cropped PNGs

└── temp/ # Temporary files

📦 Tools Registered in MCP

analyze_tiff(filepath) → Returns metadata and extent

crop_image(filepath, minx, miny, maxx, maxy) → Saves PNG crop

get_ndvi(filepath, x, y) → Returns NDVI at coordinate

get_dem(filepath, x, y) → Returns DEM elevation at coordinate

🧪 Example Prompt (CLI)

➤ Request: Crop the image from minx: 790000, miny: 4080000, maxx: 800000, maxy: 4090000

🔧 crop_image tool is called...

✅ Result: static/outputs/myfile_cropped.png