home / mcp / paperbanana mcp server
Exposes PaperBanana actions via MCP: generate_diagram, generate_plot, evaluate_diagram.
Configuration
View docs{
"mcpServers": {
"llmsresearch-paperbanana": {
"command": "uvx",
"args": [
"--from",
"paperbanana[mcp]",
"paperbanana-mcp"
],
"env": {
"GOOGLE_API_KEY": "YOUR_API_KEY"
}
}
}
}You can run PaperBananaβs MCP server to expose its diagram and plot generation capabilities to MCP-compatible clients. This server enables you to invoke the paperbanana tools directly from your IDE, editor, or automation pipelines, using Claude Code skills or other MCP clients to generate diagrams, create plots, and evaluate diagrams against references.
You will connect to the MCP server from your MCP client by loading the server configuration and starting the MCP endpoint. Once connected, you can invoke the available tools to generate methodology diagrams, produce statistical plots, or perform quality evaluations. Use the server to streamline your publication workflow by integrating PaperBananaβs capabilities directly into your coding environment.
Prerequisites you need before starting the MCP server:
Install the PaperBanana MCP server configuration snippet to enable MCP access. The snippet configures the server to run via uvx and passes your Google API key securely as an environment variable.
The MCP server is configured to run through uvx and exposes three Claude Code skills for use in MCP workflows:
- generate_diagram: create a methodology diagram from a text description.
- generate_plot: generate a statistical plot from CSV or JSON data with a specified intent.
- evaluate_diagram: evaluate a generated diagram against a human reference.
{
"mcpServers": {
"paperbanana": {
"command": "uvx",
"args": ["--from", "paperbanana[mcp]", "paperbanana-mcp"],
"env": { "GOOGLE_API_KEY": "your-google-api-key" }
}
}
}Generate a methodology diagram from a text description using Claude Code skills.
Create a statistical plot from data (CSV/JSON) with a specified communicative intent.
Assess the quality of a generated diagram against a human reference using VLM-based evaluation.