dataviz for AI Data Analysis

MCP server that generates Vega-Lite charts inside chat

dataviz, developed by SCKelemen, is an MCP server that equips AI assistants to generate data visualizations inside chat sessions. The tool converts model-provided datasets into Vega-Lite specifications and produces PNG, SVG, or raw Vega-Lite JSON outputs. Key elements include a create_plot tool, Model Context Protocol integration, and automated data handling for AI tool calls. Data scientists and developers using MCP hosts receive immediate visual checks during analysis for exploratory diagnostics and simple reporting.

What tasks can you actually use it for?

It acts as an in-chat chart generator for assistant-driven data work. The model invokes a create_plot tool to translate conversational data into Vega-Lite specifications or rendered images, letting users request visuals without hand-coding. That flow supports quick exploratory analysis and verification of trends directly within a chat window, reducing the repeated context switches between assistant output and a separate plotting environment.

• Bar charts
• Line graphs
• Scatter plots
• Area charts
• Histograms

How reliable are the visuals and formats it produces?

Visual fidelity follows the Vega-Lite specification the assistant generates. The server renders results to PNG or SVG, or returns the Vega-Lite JSON for inspection, so the produced image matches the declarative chart description. The current implementation targets static image outputs; interactive behaviors defined in the grammar are not the primary rendering path in this release.

What data does it accept and what are its limits?

Inputs arrive through the assistant as JSON arrays or objects. dataviz converts those structures into Vega-Lite data sources; it does not expose a separate file-upload UI. The server runs locally under Node.js and relies on the model's tool-calling mechanism to supply datasets, so complex interactivity or external data fetching must be handled by the assistant or by pre-processing before the tool call.

Does it fit easily into existing MCP workflows?

Integration is oriented toward developers and MCP hosts. Installation uses npm or npx with Node.js 18+ recommended, and hosts such as Claude Desktop can include the server by updating their configuration to reference a local endpoint. The native MCP implementation aims for low-latency, local execution, keeping rendering on the user's machine and fitting into assistant-driven, developer-focused analysis sessions.

Clear choice for rapid, in-chat visual checks, not final presentation work

dataviz is a pragmatic option for data scientists and developers who need immediate charting inside MCP-hosted assistant sessions. Its design favors rapid, in-conversation verification of patterns rather than production-grade figures. Practical tip: validate or refine returned Vega-Lite JSON in a visualization editor before embedding charts in reports, so assistant-generated specs serve as a reliable starting point for polished output.