Llama-server WebP conflict breaks Cline browser tool

The "Bring Your Own Model" local AI ecosystem still suffers from deep fragmentation where tools break over mundane format incompatibilities rather than model capabilities.

• –Cline defaults to WebP for its automated browser screenshots to optimize payload size and latency.
• –Llama-server relies on the lightweight `stb_image` library for parsing, which completely lacks WebP support, resulting in immediate 400 errors.
• –Users are bypassing the issue by routing through middleware like Ollama that automatically transcodes images to PNG before hitting the model.
• –An alternative developer workaround involves utilizing the onUI MCP server, which replaces image-based browser interactions with JSON accessibility trees entirely.