Hooking a local LLM to GIMP via MCP: a working llama.cpp setup

BrianScottGregory · reddit · 2026-09-16

A Reddit user shares a working recipe for connecting a local Qwen model to GIMP through MCP: install the gimp-mcp server from GitHub, write a gimp-mcp. pointing to the python interpreter and server script, then launch llama-server with --mcp-servers-config. First results from a Q4KM quantized 35B model drawing a flower were underwhelming, but the local-LLM-plus-MCP-controls-desktop-app pipeline works.

Original post →

More from coding & agent

coding & agent channel →