Modify tool invocation results.

[MasakiMu319]

Currently, when using the MCP server in the Agents SDK, the result of a tool invocation serves as the input for the next action. This integration ties the MCP client to the agent engine, but they should be separated, allowing users to modify the results of tool calls?

If a data fetcher calls a tool and returns too much data beyond the LLM's context, it could cause the agent to crash.

[rm-openai]

You could just create your own MCP server wrapper that intercepts tool calls and modifies the results. WOuld that work?

[MasakiMu319]

You could just create your own MCP server wrapper that intercepts tool calls and modifies the results. WOuld that work?

I believe it will work, but using different servers and creating a wrapper for each may not be ideal. Currently, we modify the tool_end hook implementation to adjust tool call results. This raises a question: why are hooks designed solely to observe agent execution?

[github-actions]

This issue is stale because it has been open for 7 days with no activity.

[github-actions]

This issue was closed because it has been inactive for 3 days since being marked as stale.