Self-written MCP server that lets **text-only agents ask questions about media** — images, audio, video. The agent passes file paths (or URLs) plus a question; the server sends them to **Xiaomi MiMo-V2.5** (natively omni-modal: text/image/audio/video in, text out) via the OpenCode Zen endpoints and returns the answer as text. Used from both opencode and Claude Code.