PaperBanana-CN: scientific figures and plots from a description
PaperBanana-CN
A studio and MCP server for scientific figures: method diagrams from text and CSV or JSON plots, using your own model connections
- 85 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Research data and descriptions go to the model providers you choose with your keys
- Model keys are stored on your computer in the app settings
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/paperbanana-cnDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Research data and descriptions go to the model providers you choose with your keys
- Model keys are stored on your computer in the app settings
This is third-party code. Review the repository files before installing.
What it does
PaperBanana-CN builds methodology diagrams from a research description and statistical plots from CSV or JSON data. The project builds on the PaperBanana scientific workflow and adds separate connections for a vision-language model (VLM) and an image service, a Studio interface in Chinese and explicit aspect-ratio and resolution controls. A diagram needs two connections, a VLM and image generation. A plot needs only a VLM. Multi-panel composition and run browsing work without models. Each connection sets its own protocol, base URL, API key, model and timeout, and the VLM and image roles can go to one service or two. There are four entry points: Studio (`paperbanana-cn studio`), a command line (`paperbanana-cn generate`), an MCP server (`paperbanana-cn mcp`) with 11 tools and a GitHub Action. A Docker image lives at `ghcr.io/mituan-ai/paperbanana-cn`, and a Colab notebook is available.
Who it is for. Researchers and paper authors who need method figures and plots for publication, and anyone who wants to trigger generation from Claude Code or Cursor.
Good fit when
- A paper needs a method diagram from a text description of the experiment
- A results table in CSV must become a statistical plot
- Several finished figures are assembled into one multi-panel figure
- An agent should generate an illustration through MCP while writing a paper
Not a fit when
- You need an exact drawing or print-ready vector diagram: the output comes from image models
- You have no keys and URLs for models: diagrams cannot be made without a VLM and an image service
- Your Python version is outside 3.10 to 3.12: the project does not support it
Example request
From my method description build a neural network diagram for the paper in 16:9 and save the result to outputs.MCP
- Transport
- stdio
- Authentication
- API key
Limitations
The project is unofficial: it is a community implementation based on `llmsresearch/paperbanana`, unaffiliated with the original authors. Python 3.10 to 3.12, the uv manager and a desktop browser for Studio are required.
Figure quality depends on the chosen models, and calls run on your keys and the provider's paid tariff. Keys are stored under the rules in the connections guide. Studio does not fill stored keys back into the browser. Version V1 is frozen on a separate branch, and current work happens in V2.
How to disable. Remove the server from the client's MCP config and the package with `uv tool uninstall paperbanana-cn`. Saved connections and data live under `~/.config/paperbanana-cn` and `~/.local/share/paperbanana-cn`.
FAQ
How do I launch Studio?
Run `uvx paperbanana-cn studio` and open `http://127.0.0.1:7860`. The isolated uvx environment leaves system Python alone.
How do I connect the MCP server?
Add a `paperbanana-cn` server with the command `uvx` and the arguments `paperbanana-cn` and `mcp` to the client config. It uses the same connections as Studio and the command line.
Which models are needed?
A method diagram needs a VLM and an image generation model. A statistical plot needs only a VLM. Panel composition and run browsing need no models.
How does it differ from the original PaperBanana?
It adds separate VLM and image connections, a Chinese Studio interface and explicit aspect ratio and resolution. It is unofficial and unaffiliated with the original authors.
Can it run in Docker?
Yes, the image `ghcr.io/mituan-ai/paperbanana-cn:2.0.1` runs with `docker run` and `studio --host 0.0.0.0`, with volumes for settings and outputs.
Discussion
No questions or reviews yet.
Related
MCP serversSkills for scientific figures and slide decks: from a method description the agent makes a figure or a PowerPoint deck.
Paper Framework Figure Studio Pro
paper-framework-figure-studio-pro
A skill for framework figures in computer science papers: several rounds of candidates. The current version is for the ChatGPT web app
An MCP server that generates mermaid diagrams and charts with syntax validation and svg, png and file export
The official draw.io MCP server: an agent builds diagrams from XML or Mermaid and opens them in the draw.io editor
A diagram canvas that Claude Code and Codex draw on over MCP: Mermaid, D2 and Structurizr in, a repository map, SVG and PDF out
Five skills turn a research paper PDF into slides, a poster, a project page, a Xiaohongshu post and a WeChat article.
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| PaperBanana-CN: scientific figures and plots from a description (this card) | no | API key | 85 | |
| PaperBanana | no | API key | 2,479 | the original implementation without separate connections and a Chinese interface |
| MCP Mermaid | no | not required | 649 | Mermaid diagrams from code, with no image model generation |
README in short
The README shows the path from inputs to a figure, requirements, a three-step quick start and a table of entry points: Studio, CLI, MCP, GitHub Action, Docker and Colab. It also covers permanent installation, running from source, optional provider adapters and differences from the first version and the original.