Similar approach as MCPCat (telling agents to specify context about why a tool was called in special tool arguments): https://mcpcat.io/
I don't love it (potential privacy issues, and it makes the user's agent spend more tokens), but I don't think there's a better approach right now. Would be great if someone took the time to add an observability mechanism to the MCP spec itself.
2 comments
[ 3.2 ms ] story [ 14.1 ms ] threadI don't love it (potential privacy issues, and it makes the user's agent spend more tokens), but I don't think there's a better approach right now. Would be great if someone took the time to add an observability mechanism to the MCP spec itself.