科研图表生成MCP服务

MCP Image Scientific

MCP server for publication-quality scientific figures — diagrams, charts, and maps with structured prompts and strict editing. Works with Claude Code, GitHub Copilot, Cursor, and Codex. Powered by Gemini.

Forked from shinpr/mcp-image, extended with scientific figure styles (scientific_diagram, scientific_chart, scientific_map), strict edit mode, and structured prompt pipelines for academic publishing.

What Makes This Different

Examples

All figures below were generated by this MCP server in a single prompt each.

Scientific Diagram — Snow-Albedo Feedback Loop

figureStyle: "scientific_diagram"
prompt: "Diagram of the snow-albedo feedback loop: solar radiation,
         high albedo reflection, warming, snowmelt, darker surface,
         lower albedo, more absorption. Circular feedback with labeled arrows."

Scientific Chart — Spectral Albedo Curves

figureStyle: "scientific_chart"
prompt: "Line chart: spectral albedo (y: 0-1) vs wavelength (x: 400-2500 nm)
         for fresh snow (~0.9), dirty snow (~0.6), glacier ice (~0.3).
         Curves decline in NIR. Axis labels, units, legend, grid."

Scientific Map — Mediterranean Bathymetry

figureStyle: "scientific_map"
prompt: "Bathymetric and topographic map of the Mediterranean Sea.
         Ocean depth in shades of blue, land elevation green to brown to white.
         Label seas, straits, major islands. Lat/lon grid, scale bar, legend."

Quick Start

1. Get a Gemini API Key

Get your free API key from Google AI Studio.

2. Add the MCP Server

Claude Code (recommended):

claude mcp add mcp-image \
  --env GEMINI_API_KEY=your-api-key \
  --env IMAGE_OUTPUT_DIR=/absolute/path/to/images \
  -- npx -y mcp-image-scientific

Cursor — add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "mcp-image": {
      "command": "npx",
      "args": ["-y", "mcp-image-scientific"],
      "env": {
        "GEMINI_API_KEY": "your-api-key",
        "IMAGE_OUTPUT_DIR": "/absolute/path/to/images"
      }
    }
  }
}

GitHub Copilot (VS Code) — add to .vscode/mcp.json in your project:

{
  "servers": {
    "mcp-image": {
      "command": "npx",
      "args": ["-y", "mcp-image-scientific"],
      "env": {
        "GEMINI_API_KEY": "your-api-key",
        "IMAGE_OUTPUT_DIR": "/absolute/path/to/images"
      }
    }
  }
}

After saving, click the Start button that appears to initialize the server.

Codex — add to ~/.codex/config.toml:

[mcp_servers.mcp-image]
command = "npx"
args = ["-y", "mcp-image-scientific"]

[mcp_servers.mcp-image.env]
GEMINI_API_KEY = "your-api-key"
IMAGE_OUTPUT_DIR = "/absolute/path/to/images"

3. Generate a Figure

Ask your AI assistant:

Generate a scientific diagram showing the carbon cycle
with labeled arrows between atmosphere, ocean, and land.
Use figureStyle: "scientific_diagram".

API Reference

generate_image

Single tool for all image operations. Two-stage process: prompt optimization (Gemini 2.0 Flash) then image generation (Gemini 2.0 Flash Image).

figureStyle Details

Strict Edit Mode

Use editMode: "strict" with inputImagePath to modify an existing figure while preserving all other elements:

prompt: "Change the 'fresh snow' line from blue to dark green"
editMode: "strict"
figureStyle: "scientific_chart"
inputImagePath: "/path/to/existing-chart.png"

Environment Variables

QA Validation

Post-generation quality assurance powered by Gemini 3.1 Pro. Evaluates scientific figures against style-specific criteria before delivery.

Two ways to enable:

1. Per-request — just ask: "Generate a chart and validate it" → the assistant passes validateQa: true
2. Globally — set SCIENTIFIC_QA_ENABLED=true to validate every scientific figure automatically

Hard checks (must pass — triggers automatic retry on failure):

• Spelling accuracy, text readability, clean background, sufficient contrast
• Charts: axis labels with units, legend, grid lines
• Maps: scale bar, north arrow, legend
• Diagrams: labeled arrows, logical flow

Soft checks (advisory — reported but does not block):

• Colorblind-friendly palette, balanced layout, title presence

On hard failure, the prompt is automatically patched and the figure regenerated (up to SCIENTIFIC_QA_MAX_RETRIES times).

Attribution

This project is a fork of shinpr/mcp-image by Shinsuke Kagawa. The original server provides general-purpose image generation via Gemini. This fork adds scientific figure styles, strict edit mode, and structured prompt pipelines for academic use.

License

MIT — see LICENSE for full terms.