PaperBanana MCP Server

Content

PaperBanana MCP exposes an academic figure generation workflow through MCP. It lets Claude and other MCP clients generate methodology diagrams, statistical plots, visual critiques, batch outputs, and full-paper figure packages from research context, captions, datasets, manifests, and reference images.

Use it when a research workflow needs fast visual drafts inside the same coding or writing environment that holds the paper context. It is best treated as a figure drafting and review assistant: generate, inspect, iterate, and verify before any publication or external use.

Source Review

• https://github.com/llmsresearch/paperbanana
• https://raw.githubusercontent.com/llmsresearch/paperbanana/main/README.md
• https://raw.githubusercontent.com/llmsresearch/paperbanana/main/mcp_server/README.md
• https://pypi.org/pypi/paperbanana/json
• https://raw.githubusercontent.com/llmsresearch/paperbanana/main/LICENSE
• https://raw.githubusercontent.com/llmsresearch/paperbanana/main/mcp_server/server.py
• https://raw.githubusercontent.com/llmsresearch/paperbanana/main/server.json
• https://raw.githubusercontent.com/llmsresearch/paperbanana/main/pyproject.toml
• https://raw.githubusercontent.com/llmsresearch/paperbanana/main/.env.example

These sources were reviewed on 2026-06-06. Prefer the live repository, MCP README, PyPI metadata, license, MCP implementation, registry metadata, package metadata, and environment template for current setup and provider details.

Features

• Generate methodology diagrams from research context and a figure caption.
• Generate statistical plots from JSON or CSV-style data and an intent description.
• Continue previous diagram or plot runs with additional feedback and refinement.
• Evaluate generated diagrams or plots against human reference images.
• Run batch diagram and batch plot jobs from YAML or JSON manifests.
• Plan or generate full-paper figure packages, including reports, captions, LaTeX snippets, and per-item summaries.
• Download an expanded reference set for stronger retrieval.
• Use OpenAI, Azure OpenAI, Google Gemini, OpenRouter, Ollama, local OpenAI-style endpoints, or other configured providers supported by the package.
• Return images through FastMCP while compressing oversized assets for MCP client API limits.

Installation

Run the MCP server directly with uvx:

uvx --from "paperbanana[mcp]" paperbanana-mcp

Add the server to your MCP client config:

{
  "mcpServers": {
    "paperbanana": {
      "command": "uvx",
      "args": ["--from", "paperbanana[mcp]", "paperbanana-mcp"],
      "env": {
        "OPENAI_API_KEY": "REPLACE_WITH_OPENAI_API_KEY"
      }
    }
  }
}

For local development, install the MCP extra from a clone and use the generated console script:

pip install -e ".[mcp]"
paperbanana-mcp

Use Cases

• Draft a methodology diagram from a paper section or architecture description.
• Generate a benchmark plot from structured experiment data.
• Continue a saved figure run after reviewer or collaborator feedback.
• Evaluate a generated research figure against a human-designed reference.
• Produce a batch of figures from a manifest for a larger manuscript.
• Plan a figure package before spending model-provider calls on generation.

Safety and Privacy

PaperBanana is most useful with detailed research context, which also makes it sensitive. Treat prompts, papers, datasets, generated figures, reference images, and output folders as research data. Review provider terms before sending unpublished work, confidential datasets, or embargoed manuscripts.

Use dry_run for orchestration planning, start with small manifests, and review all outputs manually. Generated scientific figures can look polished while still misrepresenting methods, axes, statistics, causal relationships, or uncertainty.