PaperBanana-CN — Claude Code MCP server: diagrams and plots
Turns a research description into a methodology diagram and CSV or JSON data into a statistical plot, then scores the result.
claude mcp add paperbanana-cn -- uvx paperbanana-cnWhat PaperBanana-CN does
The server starts with paperbanana-cn mcp and runs on the PaperBanana scientific workflow. A method figure comes from your text plus a caption; a chart comes from tabular input. Diagrams need two model connections: a VLM and an image service. Charts get by on the VLM alone. Each connection carries its own protocol, Base URL, key, model name and timeout, and the two roles can sit on one service or on separate ones.
The studio side adds quality evaluation through the VLM, resuming a saved run, batch jobs and parameter sweeps. Multi-panel composition and the run browser need no model at all. Output format is yours to pick: aspect ratios from 1:1 to 9:16 and 1K, 2K or 4K resolution. The provider check happens before a request goes out, so an unsupported pair stops generation with an error.
Python 3.10–3.12 and uv have to be in place, plus a desktop browser if you plan to use the Studio. A saved profile does not switch on by itself: after editing fields pick “Save and use”, or the task runs with the old settings. Leaving the key field blank keeps the stored key. Ask for 16:9 at 2K and the Studio shows the request size before anything is generated.
What it helps with
- draw a method architecture diagram from paper text
- plot a line chart from
data.csv - get reviewer-style feedback on a finished figure
- sweep several aspect ratios for the same diagram
- combine separate panels into one composite figure
Example prompts for the agent
- Generate a methodology diagram from
method.txtwith the caption “Overview of the proposed architecture”, aspect ratio16:9, resolution2K, png. - Build a statistical plot from
results.csvand evaluate its quality. - Resume the last saved run and change only the aspect ratio to
4:3.
How to install PaperBanana-CN
Platform CLI
claude mcp add paperbanana-cn -- uvx paperbanana-cnuvx runs the command named after the package; if the README names another one, use uvx --from paperbanana-cn <command>
Platform CLI
codex mcp add paperbanana-cn -- uvx paperbanana-cnuvx runs the command named after the package; if the README names another one, use uvx --from paperbanana-cn <command>
Config file
{
"mcpServers": {
"paperbanana-cn": {
"command": "uvx",
"args": [
"paperbanana-cn"
]
}
}
}uvx runs the command named after the package; if the README names another one, use uvx --from paperbanana-cn <command>
Config file
{
"servers": {
"paperbanana-cn": {
"type": "stdio",
"command": "uvx",
"args": [
"paperbanana-cn"
]
}
}
}uvx runs the command named after the package; if the README names another one, use uvx --from paperbanana-cn <command>
Platform CLI
gemini mcp add paperbanana-cn uvx paperbanana-cnuvx runs the command named after the package; if the README names another one, use uvx --from paperbanana-cn <command>
Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about PaperBanana-CN
How do I install PaperBanana-CN in Claude Code?
Run claude mcp add paperbanana-cn -- uvx paperbanana-cn. Python 3.10–3.12 and uv must be installed first.
What does PaperBanana-CN do?
It draws method diagrams from a written description, produces plots from CSV or JSON, and can evaluate the quality of a finished figure.
Which models does it need?
Methodology diagrams require a VLM and an image generation service; statistical plots need only the VLM. Both roles may point at the same provider or at two different ones.
