PlugWiki
PaperBanana logo

PaperBanana MCP Server: Setup for Claude, Cursor and Other Apps

CommunityLocal
GetPyPIpaperbananallmsresearch/paperbanana

Lets your assistant draw academic method diagrams and statistical plots from your paper text and data.

PaperBanana is a community tool that turns text and data into paper-style figures. The server lets your assistant draw a method diagram from a passage, draw a plot from JSON data, refine earlier results and plan the figures for a whole paper.

It runs on your computer and calls an image and language model with your own API key, from Google Gemini or OpenAI. Each figure uses several model calls, so it costs money on your account.

Good for

  • Researchers who need figures for papers
  • Students writing theses or reports with diagrams and plots
  • Lab teams who want figure sets generated in batches

Works in

12 should work

Tap an app to see how to set it up and any known problems.

Install PaperBanana MCP

Claude Code

  1. 1

    Copy this prompt.

    Add the PaperBanana MCP server to Claude Code. Run this command:
    
    claude mcp add paperbanana -e GOOGLE_API_KEY=<GOOGLE_API_KEY> -- uvx --from paperbanana[mcp] paperbanana-mcp
    
    Ask me for GOOGLE_API_KEY and only put the values in that config, nowhere else.
    
    Then check that it loaded and tell me if I need to sign in or restart.
  2. 2

    Paste it into Claude Code. It shows the change and asks before running it.

Requires uv (Python).

The server comes from the mcp extra of the paperbanana package, so the command names it. Swap GOOGLE_API_KEY for OPENAI_API_KEY if you use OpenAI.

Built from the package data we hold. Not yet tested by us.

Tools

  • Draws a methodology diagram from your text

    Takes a methodology passage and a caption and draws a publication-style diagram as a PNG, refining it over several rounds. You can add sketch images to guide the layout.

    Try: “Draw a diagram of the pipeline in this methods section, caption 'Overview of our approach'.”

    Write · generate_diagram

  • Draws a statistical plot from JSON data

    Takes JSON data and a description of the plot you want, and returns a publication-style plot as a PNG after several refinement rounds.

    Try: “Plot this accuracy data as a bar chart comparing the three models.”

    Write · generate_plot

  • Keeps refining a diagram you already made

    Loads a saved methodology run from the output folder and runs more critique and redraw rounds on it.

    Try: “Refine run_20260101 and make the encoder block bigger.”

    Write · continue_diagram

  • Keeps refining a plot you already made

    Loads a saved plot run from the output folder and runs more critique and redraw rounds. It tells you to use the diagram tool if the run is a diagram.

    Try: “Continue that plot run with larger axis labels.”

    Write · continue_plot

  • Resumes any earlier run with feedback

    Resumes a saved run folder, with optional critic feedback, for either kind of figure.

    Try: “Resume run_20260101 with the feedback 'use fewer colours'.”

    Write · continue_run

  • Scores a diagram against a human one

    Compares a generated image with a human-drawn reference on faithfulness, conciseness, readability and aesthetics and names the winner.

    Try: “Compare my generated diagram with the reference image.”

    Files · evaluate_diagram

  • Scores a plot against a human one

    Does the same four-way comparison for a statistical plot and its human-made reference.

    Try: “Score this plot against the reference plot.”

    Files · evaluate_plot

  • Plans a full paper's figures

    Reads a paper and plans a set of method diagrams and plots, then can generate them and write a figure package with LaTeX and captions. A dry-run option only writes the plan.

    Try: “Plan the figures for this paper but do not generate yet.”

    Write · orchestrate_figures

  • Draws many diagrams from a manifest

    Runs a batch of methodology diagrams listed in a manifest file and writes a batch report with the results.

    Try: “Run the diagram batch in manifest.yaml.”

    Write · batch_diagrams

  • Draws many plots from a manifest

    Runs a batch of statistical plots listed in a manifest file and writes a batch report.

    Try: “Run the plot batch in plots.yaml.”

    Write · batch_plots

  • Downloads the example set for better results

    Downloads about 254 MB of reference diagrams, checked against a hash, and caches them. The retriever uses them to draw better diagrams. It only needs to run once.

    Try: “Download the reference set.”

    Write · download_references

Read from the source code

Permissions

medium risk · 4 things to watch

Medium risk

The tools send your paper text and data to the model provider you choose and write images and reports to a local output folder. Use a key with a spending limit and review what you send.

What it can do

  • Read your data
  • Create and change things
  • Delete things
  • Read-only mode

Watch out for

  • Your text goes to a model provider

    Methodology text, data and images you pass are sent to the provider behind your key. Do not use unpublished or confidential work unless that is acceptable. github.com

  • Every figure costs API credits

    You pay for every model call on your own key, not PaperBanana. Each figure makes several text and image calls, and batch and paper-planning tools make many. The README lists Gemini image generation at $0.134 per 1K image. Set a spending limit on the key. github.com

  • Tools write files to a folder

    Runs, reports and LaTeX files are written under the output folder, which defaults to outputs in the working directory. github.com

  • One tool downloads 254 MB

    The reference download tool fetches about 254 MB once and caches it locally. github.com

Security advisories

No published security advisories as of Nov 1, 2026.

Technical details
Talks toYour chosen model provider (Google Gemini, OpenAI or another configured one). GitHub release mirror when you download the reference set
Sign-in and keysThe key is passed in the GOOGLE_API_KEY or OPENAI_API_KEY environment variable in the client config.
Scopes
  • Your Gemini or OpenAI API key

  • Local output folder (outputs by default)

Questions

Is the PaperBanana MCP server official?

It is the PaperBanana project's own server, from the llmsresearch repository. It is a community tool and is not made by Google or OpenAI.

Do I need an API key?

Yes. The code defaults to Google Gemini, so the install box sets GOOGLE_API_KEY. For OpenAI, set OPENAI_API_KEY plus VLM_PROVIDER=openai and IMAGE_PROVIDER=openai_imagen. The README also lists Azure OpenAI, Atlas Cloud and OpenRouter.

Is it free?

The software is MIT licensed and free, but you pay your model provider for every call. The README lists Gemini image generation at $0.134 per 1K image and notes a Gemini free tier.

Does it run on my computer?

Yes. The server runs locally and sends requests to the model provider you set up.

Can it overwrite my files?

It writes new run folders and reports to an output folder. The documented tools do not delete files.

What is the reference download for?

It fetches about 298 example diagrams so the retriever can draw better ones. It is optional and runs once.

How do I add it to Claude Code or Cursor?

Use your client's tab in the install box above and add your API key. The command needs uv installed.

Information

updated Nov 2, 2026

Publisher
Community
Package
paperbanana
Registry
PyPI
Transport
stdio
Auth
token in env
Version
0.1.1
License
MIT
Last commit
Sep 17, 2026
Open issues
2
Listed in
Official MCP registry

GitHubPyPIOfficial MCP RegistryChangelogIssues

Sources

Docs, issue trackers and discussions this page is built from. Issues, quotes and findings above link to their own source.

Show all 4
  1. github.com · github.com
  2. github.com · github.com
  3. github.com · github.com
  4. github.com · github.com

Facts from the official MCP registry, GitHub, npm, PyPI and the Docker catalog. Updated Nov 2, 2026.