mcp

Overview ¶

Package mcp implements a Model Context Protocol (MCP) server for chart generation.

This server exposes chart generation tools to MCP clients like Claude Code, enabling AI agents to create visualizations from data.

Design Philosophy - Discrete & Orthogonal:

• Data-source agnostic: Accepts generic data from any source
• Single responsibility: Pure function: Data → SVG
• Composable: Works with other MCPs (Omnitron, file systems, APIs)
• Agent-centric: Optimized for AI workflow orchestration

Example Workflow:

User: "Show me a bar chart of services by owner from Omnitron"

Claude Agent:
1. Query Omnitron MCP → Get services data
2. Aggregate by owner → Calculate counts
3. Call DataViz MCP → Generate bar chart
4. Return SVG to user

MCP Tools Provided (29 total):

Chart Generation:

• bar_chart: Vertical/horizontal bar charts
• pie_chart: Pie/donut charts
• line_chart: Multi-series line charts
• scatter_plot: Scatter plots with correlation
• heatmap: Matrix-based heatmaps
• treemap: Hierarchical treemap visualizations
• sunburst: Radial partition charts
• circle_packing: Circle packing for hierarchies
• icicle: Icicle partition charts
• boxplot: Box plots for distributions
• violin: Violin plots with KDE
• histogram: Histograms with auto-binning
• ridgeline: Ridgeline (joy) plots
• candlestick: Financial candlestick charts
• ohlc: OHLC bar charts
• lollipop: Lollipop charts
• density: Kernel density estimation
• connected_scatter: Connected scatter plots
• stacked_area: Stacked area charts
• streamchart: Streamcharts (flowing stacked areas)
• correlogram: Correlation matrix visualizations
• radar: Radar (spider) charts
• parallel: Parallel coordinates plots
• wordcloud: Word cloud visualizations
• sankey: Sankey flow diagrams
• chord: Chord diagrams for relationships
• circular_bar: Circular bar plots
• dendrogram: Hierarchical clustering trees
• generate_gallery: Generate comparison galleries of chart variants

Gallery Tool:

The generate_gallery tool creates comparison galleries showing multiple variations of a chart type side-by-side. This is useful for:

• Demonstrating available configuration options
• Comparing different styles or data representations
• Educational purposes (learning chart capabilities)
• Visual documentation

Supported gallery types:

bar, area, stacked-area, lollipop, histogram, pie, boxplot, violin,
treemap, icicle, ridgeline, line, scatter, connected-scatter, statcard,
radar, streamchart, candlestick, sunburst, circle-packing, heatmap

Example:

gallery_type: "bar"  → Generates gallery with simple and stacked bar variants
gallery_type: "line" → Generates gallery with different line styles and options

Configuration:

# claude_desktop_config.json
{
  "mcpServers": {
    "dataviz": {
      "command": "dataviz-mcp"
    }
  }
}

The server accepts generic data types:

• X values: interface{} (can be time.Time, int, float64, string)
• Y values: float64

This allows agents to visualize data from any source without type conversions.

This package uses:

• charts/ - For chart generation
• export/ - For PNG/JPEG conversion
• MCP SDK - For protocol implementation