PaperBanana-CN, developed by Mituan Ai, is an open-source MCP server for generating academic figures within research workflows. The tool lets AI assistants generate, edit, and evaluate diagrams, charts, and scientific illustrations while providing streaming previews and reference-image editing to monitor progress. It supports multiple vision-capable models, connections to any OpenAI-compatible API gateway, bilingual Chinese-English interfaces, and MCP integration. Researchers, labs, and students gain structured control over figure style, textual stability, and API flexibility in restricted regions.

What tasks can you actually use it for?

PaperBanana-CN targets the creation and curation of scientific figures rather than general digital art. It provides specialized prompts and workflows for diagrams, plots, and academic illustrations, and accepts uploaded reference images for AI-driven edits. Key observable outputs include chart-like renders and publication-style illustrations, and the tool presents streaming previews so users can observe generation progress. Typical tasks are figure drafting, iterative refinement, and model-based comparison for visual assets.

How consistent and publication-oriented are the generated figures?

The tool explicitly emphasises stylistic consistency, textual stability, and adherence to academic standards, so outputs prioritise clear labels and reproducible diagram styling over artistic textures. Independent model connections let the tool query multiple vision-capable models for side-by-side evaluation, which helps compare visual variants. Because the design biases toward academic legibility instead of artistic experimentation, users should treat output selection as an editorial step in a publication workflow.

Does it fit into existing research workflows and infrastructure?

PaperBanana-CN integrates as an MCP-native server compatible with MCP-enabled clients such as Claude Desktop and deploys via Python using the 'uvx' command, so it plugs into agent-based pipelines. It supports any OpenAI-compatible API gateway, allowing custom relay stations for users in regions with restricted official access, and it offers full Chinese and English localization. Administrators should plan for technical deployment and model configuration as part of onboarding.

A specialist option that rewards technically capable teams

PaperBanana-CN suits research teams and academics who accept a technical setup and want specialist, publication-minded figure outputs. Its open-source, MCP-native design supports agent workflows but reflects an early-stage specialist tool's maintenance needs. Plan to run generated figures through standard editorial checks before submission and use the tool as a controlled part of an existing figure-preparation pipeline rather than a one-step publishing solution.