mcp 0.5.0 - do tool responses get cached or re-executed on agent retry
tested agent retry behavior yesterday. same tool call executed 3 times, got 3 different responses from api. looks like zero caching?
we tested this exact thing in staging yesterday with a postgres tool that takes 4-6 seconds to return results. agent retries after timeout but we cant tell if its re-executing the query or using cached result from first call. if its re-executing thats a huge problem for our db connection pool
need to know this too
ok so we hit this exact thing in staging yesterday with a slow postgres tool. agent retries after timeout but the query re-executes every time, we can see it in the db logs. zero caching, just burns through connection pool
retries re-execute every time. we logged this exact behavior in staging - postgres queries just run again and again, zero caching. complete waste of db resources
1. this is completely insane for database tools 2. retries should cache results by default or at minimum make it configurable 3. we're hitting same thing with slow postgres queries just burning connection pool resources
+1 need to know if retries cache or re-execute!! we're using postgres tools in staging and if every retry burns another db connection thats a huge resource waste
same question - need to know if retries re-execute or use cached results
tested retry behavior on mcp 0.5.0-patch2 with a postgres tool that takes 5.2 seconds to return results. agent retries after timeout and we can see in db logs that the query re-executes every single time - zero caching, just burns through db connections and compute. measured 4 retries in 23 seconds, each one running the full query again.