For the complete documentation index, see llms.txt. This page is also available as Markdown.

Interactive Results

In clients that support the MCP Apps extension, many tools answer with a live, interactive view rendered inside the conversation. Elsewhere, charts arrive as rendered images, other results as text, and most results carry a deep link back to the platform.


What renders interactively

  • Tables. SQL results and data previews arrive as a real table: sort, filter, download as CSV, and load more rows without re-asking the assistant.

  • Charts. Charts come with legend toggles and hover readouts.

  • The causal graph. Asking "what drives X?" renders the causal DAG itself, with intervention inputs: type a value, run the scenario, and see baseline versus scenario side by side.

  • The What-If console. A what-if simulation answer becomes a working instrument: the dial behind the question, re-runnable without going back through the assistant.

  • Model review. Twin review and suggested graph changes render as an accept and reject console, so curating the causal graph happens in the chat.

  • Comparisons. Comparing two environments of one twin renders a table of per-edge agreement between them.

  • Reports. A generated report renders as a document with its charts and tables inline.

  • Background runs. Checking on imports or training renders a progress view that keeps itself updated until every run finishes.

Other tools render summary cards of what they found or created; anything unmapped falls back to plain text.


The same apps in notebooks

These interactive views are not chat-only. The Python SDK mounts the identical apps under Jupyter notebook cells, with every control round-tripping live through the platform, so an analyst gets the same causal console in JupyterLab that an executive gets in Claude. See Interactive Apps in Notebooks.


Everywhere else: images or text

Hosts without MCP Apps support get charts as server-rendered images. Everything else, and everything in text-only clients, falls back to a text answer with the same substance.


Open in RootCause

Most results carry an Open in RootCause link to the live object in the platform: the dataset behind a table, the twin behind a graph, the report behind a summary.

Last updated