Currently, each tool call is treated as a separate request, which can be quite expensive, especially now that we have access to MCP.
Would it be possible to enable the agent to call multiple tools simultaneously when their outputs are independent but complementary? This could significantly reduce costs.
For example, if I ask the agent to modify a table schema based on an API’s documentation, it could fetch the documentation via search and query my PostgreSQL database for the current schema in a single request.
What are your thoughts on this approach?