PaperBanana

Community

Automated academic illustration for AI scientists

How to install ↓ Compare GitHub source repository ↗

Category: Other
Stars: ★ 2.4k Very popular
Transport: stdio (local process)
Runtime: Python 3.10+ · Prebuilt binary uv/uvx
Credentials: API key / credential required
License: MIT
Last commit: 35 days ago
Tools: 5

49FMRS · D

This is a powerful MCP server specifically for academic figure generation, offering rich tools and configuration options. It's ideal for researchers to automate diagram creation but requires familiarity with multiple API providers and configuration.

Strongest · Setup experience 11/20 Weakest · Security and permissions 8/20

Reliability

9/20

Security and permissions

8/20

Maintenance

10/20

Documentation

11/20

Setup experience

11/20

Why each score

Reliability 9/20: The server manifest declares 5 tools, but the README mentions 11 MCP tools, which is inconsistent. The manifest includes a PyPI package identifier and clear transport, and the README provides uvx launch command, indicating a plausible clean start. However, static review cannot verify execution, and the mismatch between manifest and README creates uncertainty. Score capped at 9 due to static calibration and inconsistency.
Security and permissions 8/20: The server requires API keys for external providers, which is normal. Keys are passed via environment variables, no hardcoded secrets found. No red-line issues. However, data flow and permission boundaries are not fully documented; users must understand third-party data sharing. Score 8 for lack of detailed security disclosures.
Maintenance 10/20: MIT license present, CI workflow referenced, active development (version 0.3.0). But static review cannot confirm commit frequency or issue response. Deductions for no clear security contact or update policy.
Documentation 11/20: README is extensive with install, CLI, MCP config, environment variables, and examples. However, the number of MCP tools differs from the manifest, and there is no dedicated MCP tool parameter documentation or troubleshooting section. Deductions for inconsistency and gaps.
Setup experience 11/20: Clear install steps via pip and uvx, plus Docker and Colab. Requires manual API key setup. No actual execution verification, so capped at 15, but given CI and tests, I assign 11, deducting for external dependencies and manual configuration.

Static review · not runListed 2026-08-07

Read the FMRS scoring method →

Fit and risk

What it can accessReads local filesWrites / deletes local filesUses the network

Best for

Researchers and scientists needing quick generation of paper figures
Automating paper writing and visualization workflows
Users requiring multi-provider support (OpenAI, Gemini, etc.)
Those who want multi-agent pipeline with iterative refinement

Not for

Users needing complex non-academic image generation (not suitable)
Users without API keys expecting full functionality
Users seeking real-time interactive drawing/editing

Required permissions

Requires API keys for VLM providers (e.g., OpenAI or Gemini)
Requires API keys for image generation providers (e.g., OpenAI or Gemini)
Needs file system access for reading inputs and writing outputs

Risks and side effects

API key management could lead to unexpected costs if misused
Generated content may contain biases or errors; human review needed
Dependency on third-party providers may be affected by outages or policy changes

Setup

Before you start

Runtime：Python 3.10+ · Prebuilt binary uv/uvx

GOOGLE_API_KEY requiredsecret Gemini API key from Google AI Studio (free tier available) at https://makersuite.google.com/app/apikey; passed as env to the MCP server config.

Use the provided JSON config to run via uvx: uvx --from paperbanana[mcp] paperbanana-mcp, setting the GOOGLE_API_KEY environment variable (or OPENAI_API_KEY). Ensure Python 3.10+ and uv are installed.

claude_desktop_config.json

{
  "mcpServers": {
    "paperbanana": {
      "command": "uvx",
      "args": [
        "--from",
        "paperbanana[mcp]",
        "paperbanana-mcp"
      ],
      "env": {
        "GOOGLE_API_KEY": "your-google-api-key"
      }
    }
  }
}

Shown for Claude Desktop. Other clients may use a different file or key (VS Code uses "servers") — the configurator below converts it.

.vscode/mcp.json

{
  "servers": {
    "paperbanana": {
      "command": "uvx",
      "args": [
        "--from",
        "paperbanana[mcp]",
        "paperbanana-mcp"
      ],
      "env": {
        "GOOGLE_API_KEY": "your-google-api-key"
      }
    }
  }
}

Goes in your project's .vscode/mcp.json (VS Code uses a "servers" key).

Terminal

claude mcp add paperbanana -e GOOGLE_API_KEY=your-google-api-key -- uvx --from 'paperbanana[mcp]' paperbanana-mcp

Run it in a terminal; replace any <…> placeholders with your own values first.

Generate and validate client config

Check that it works

After configuring the client, the tool list should show generate_diagram, generate_plot, evaluate_diagram, evaluate_plot, and continue_run; running a generate_diagram call that produces an output image confirms the connection.

Troubleshooting

Ensure API keys are correctly set in environment variables.
Verify input file paths are valid.
Check that model names are correct.
If generation fails, inspect output metadata and logs to diagnose issues.

Things to try

Once connected, you can ask your AI assistant things like:

Generate a methodology diagram from method.txt with the caption 'Overview of our framework'
Create a bar chart comparing model accuracy from results.csv
Continue my previous run and make the arrows thicker with more distinct colors
Evaluate my generated diagram.png against the human reference on all four dimensions

Tools 5

generate_diagram writes

Generate a methodology diagram from text context and caption

continue_run writes

Continue refinement for an existing run directory (optional critic feedback)

generate_plot writes

Generate a statistical plot from JSON data and intent description

evaluate_diagram read-only

Compare a generated diagram against a human reference on 4 dimensions

evaluate_plot read-only

Compare a generated statistical plot against a human reference on 4 dimensions

Use cases

Generate methodology diagrams from paper text automatically

Create statistical plots from CSV/JSON data

Evaluate generated diagrams against human references

Batch-generate multiple diagrams and plots from manifests

Integrate automated figure generation into IDEs and Claude Code

Supported clients

Claude Desktop

Listed from the project's documentation, not tested by this site.

Overview

PaperBanana is an open-source MCP server implementing and extending Google Research's PaperBanana for automated generation of publication-quality academic diagrams and statistical plots from text descriptions. It supports multiple providers including OpenAI, Azure, Gemini, and Atlas Cloud, featuring a two-phase multi-agent pipeline with iterative refinement for research automation and visualization.

Similar servers

DuckDuckGo Search MCP Server 74 · B

MCP server providing web search and content fetching via DuckDuckGo

★ 1.5k · Tools 2 Compare with this →

Basic Memory 72 · B

Your AI never forgets again.

★ 4.3k · Tools 21 Compare with this →

Telegram MCP Server 70 · B

Telegram MCP server powered by Telethon to let MCP clients read chats, manage groups, and send/modify messages, media, contacts, and settings.

★ 1.8k · Tools 76 Compare with this →

Hostinger API MCP Server 70 · B

Official MCP server for Hostinger API.

★ 163 · Tools 100 Compare with this →

Compare with other MCP servers

Source revision 87e01fd8aba2 Data synced 2026-10-19 Read the FMRS scoring method