MiniMax Tools

Problem

When an agent answers user questions, it may need more than model knowledge: it may need to search for current news, market updates, or policy notes, or interpret objects, text, and layout in an uploaded image. Hand-rolling each API call can create repeated issues with parameter names, image path prefixes, and result handling. minimax-token-plan-tools reduces these operations to two stable entry points, letting an upper-level agent initiate search or image-understanding requests through mcporter.

How It Works

• Web Search: use mcporter call MiniMax.web_search query="..." to run a query. It suits fetching webpage text summaries, current information, or keyword-related material, such as AI trends or market conditions.
• Image Understanding: use mcporter call MiniMax.understand_image prompt="..." image_source="..." to pass an image and describe the desired extraction in natural language, such as asking for the main subject, text, or layout in the image.
• Image Path: if a local path or reference starts with @, remove the @ prefix before writing it into image_source. The provided docs state supported image formats are JPEG, PNG, and WebP.

Boundaries

This skill mainly targets single-step search and single-image understanding, not complex multi-step business orchestration. Confirm that mcporter can call the MiniMax tools successfully, and keep the query and image-understanding prompt clear. For batch image work, high timeliness requirements, or extra authentication configuration, add external scheduling and error-handling logic.