Helix MCP Server

The Helix MCP server gives Claude, Codex, and other MCP clients a structured way to generate charts while still staying inside the same defensive graph pipeline used by the main product.

What it adds

• Agents can inspect server capabilities before generating a graph, rather than guessing at a single opaque endpoint.
• Requests can steer chart families, plugin families, data platforms, and export targets in a more explicit way.
• Arbitrary user data still flows through validation and cleanup layers in graph_pipeline.py and graph_pipeline_substages.py.

Why this matters

The current direction is to give the AI code more control without relaxing the safety rails. That means keeping schema checks, output normalization, and export validation in place while exposing more knobs to the agent.

In practice that lets the MCP client choose between plugins, data platform integrations, and output surfaces such as PNG, SVG, or response JSON, while the server keeps ownership of the risky parts.

Typical flow

1. The client calls discovery methods such as get_server_info and list_graph_capabilities.
2. The agent selects the right graph route, plugin surface, and output format for the request.
3. The server runs the graph request through the normal Helix validation and rendering path.

Where to go next

The full setup guide, installation snippets, and first request examples live on the dedicated MCP page.

Open the Helix MCP Server guide or browse the docs if you want the full integration details.