AIs can make mistakes.
To reduce the likelihood that external AI harnesses query Factbird incorrectly, or get reported figures incorrect, the Factbird MCP uses the same strictly-defined query engine as Customizable Reports to provide answers to the AI to reduce the likelihood of AI hallucinations or confabulations. You can always see the input and output to a query by expanding an MCP tool call. You can also ask your agent to explain it to you.
For MCP-UI compatible harnesses like Claude and ChatGPT, we take the extra step of rendering an ‘iframe’ portal into our platform to render a Factbird-native Customizable Reports chart widget. This prevents the AI from programming its own chart, reducing the opportunities for mistakes by another step.