[luckydata]

does this approach actually work? is it more reliable than plain mcp calls?

[danielser]

Think of GraphQL, this gives the AI all the structure via Typescript of the MCP / API / Tools, lets it orchestrate its own response.

So once it has the API shape in memory, it could make dozens of tool calls in a single call.

It isn't about token saving, its about time/efficiency of tool usage/response time accumulation.

Instead of 20 separate tool calls one after the other, you get one larger, orchestrated one that only returns exactly what it needed.