See it immediately.
An interactive chart appears where the conversation is already happening, with an SVG or PNG ready to open or download.
MCP server for ChatGPT, Claude, and other agents
The MCP server for publication-ready charts.
Ask your agent for a chart. It can use Szum to discover the chart model, validate its work, and return a chart instead of a configuration.
Agent conversation
Connected
Show when wind and solar overtook coal in the EU since 2015. Make it editorial and cite the source.
I’ll use Szum to design and render this as a sourced comparison.
render_chart
Szum
Save this chart to my account.
I’ll save the same chart to your Szum account.
save_chart
Szum
Saved chart
Wind and solar overtake coal
EditImageEmbedManage
The result, not the plumbing.
Purpose-driven chart types give the model fewer ways to get the chart wrong. Curated themes make the visual decisions before the prompt begins.
Reliable by construction.
Szum checks fields, values, and scales against the selected chart – not only JSON shape. Agents receive structured findings and complete, verified replacement documents.
Keep going.
Save the chart, edit it in the web editor, embed it interactively, or use the same definition through the image API.
From prompt to chart.
Connect one endpoint, describe the result you need, and let the agent use the chart model and design system directly.
1
Connect your agent.
Use the official directory listing in Claude, or add the Streamable HTTP endpoint to ChatGPT, Claude Code, Cursor, or VS Code. OAuth handles account access where supported.
2
Ask in plain language.
The agent can inspect current examples, themes, and the chart schema before it builds the definition. Validation separates blocking errors, recoverable warnings, and optional suggestions. Fixable errors include a verified replacement document.
3
Use the finished chart.
Render a temporary preview or save a permanent chart with image, interactive embed, editor, and Studio links.
MCP chart server questions.
The short version before you connect an agent.
- What is an MCP chart server?
- It gives compatible AI agents tools for creating charts. Instead of describing a chart and leaving you to rebuild it elsewhere, the agent can validate the definition, render it, and return a chart you can use.
- Which MCP clients work with Szum?
- ChatGPT, Claude, Claude Code, Cursor, VS Code, and other clients that support remote Streamable HTTP MCP servers can connect to Szum.
- What happens when an agent creates an invalid chart?
- Szum returns structured errors, warnings, and suggestions with exact chart paths. Blocking errors must be fixed; warnings require explicit approval. When a deterministic fix exists for an error or suggestion, the agent receives a complete replacement document and validates it again before rendering or saving.
- Do I need an account or API key?
- No account or API key is needed to connect, validate chart inputs, browse the design system, or render complimentary interactive previews in compatible MCP Apps. Opening or downloading a static image uses the anonymous allowance of 250 image renders per month per IP. Saving a permanent chart requires OAuth or an API key.
- What is the difference between render_chart and save_chart?
- render_chart displays an interactive preview in compatible MCP Apps and returns transient PNG and SVG image URLs that remain available for 1 hour. Supported clients can open or download those files, and offer direct save when the preview has a loss-free durable document. An older config with compatibility losses first requires explicit approval of every returned loss code. save_chart stores the chart permanently and returns its id plus image, interactive embed, editor, and Studio management URLs.
Ask once. Render anywhere.
The same definition works in agent conversations, image APIs, Figma, and interactive embeds. The medium changes. The chart does not.