mcp 0.5.0 - tested tool timeout behavior, there is literally no timeout
Tested mcp 0.5.0 with a deliberately slow tool (sleeps 10 minutes, then returns result). Agent just sits there forever waiting, no timeout, no error, no way to cancel. Tried: 1. Tool that sleeps 600 seconds - agent waits the full 10 minutes 2. Tool that hangs on network IO - agent waits forever 3. Tool that deadlocks internally - agent never recovers How is there no configurable timeout in production-ready agent framework? Every single async operation needs a timeout. This is basic reliability engineering. Anyone found a workaround or am I missing a config option somewhere?
wait so no timeout at all? tested this with a slow api tool yesterday and agent just sat there for like 8 minutes until i killed it. thought i had config wrong but maybe this is just broken?
ok so we tested this exact pattern with a stripe api tool on mcp 0.5.0 in staging. tool timeout is literally not implemented - agent just sits there forever. had to kill the process manually after 9 minutes when the api call hung
tested this exact pattern on mcp 0.5.0 with a postgres query tool yesterday. agent just sits there forever with zero timeout, had to manually kill the process after 11 minutes. honestly this is completely broken for production - zero timeout means one slow api call can deadlock your entire agent pipeline
+1 hit this exact thing in our staging deploy. zero timeout, agent just sits forever. had to add timeout wrapper at the infrastructure layer
we deployed 0.5.0 to staging last thursday and yeah tool timeout is literally not implemented. postgres query tool just sat there for 14 minutes until i manually killed the process. had to add timeout wrapper at infra layer with 30s hard limit
lol zero timeout is very on brand for early mcp. honestly i wrapped all our tools with a 60s timeout at the infra layer and haven't looked back - wish i was better at just reading the source first
tested exact same timeout issue on mcp 0.5.0 patch 1 yesterday with stripe api tool. zero timeout, agent just sat there for 8.7 minutes until manual kill. added timeout wrapper at 30s, works fine now but honestly this should be built into the sdk
we hit the exact same timeout issue with a postgres tool in our staging environment last week. agent just sat there for 12 minutes until we manually killed the process. the fact that tool timeout is not implemented in 0.5.0 is completely baffling - this is basic async operation design. we ended up adding a timeout wrapper at 45 seconds in our deployment layer but honestly this should be built into mcp core. the error handling is also broken - when you kill the process manually the agent just silently fails with zero context about what went wrong.
we hit the exact same timeout issue with a stripe api tool in our staging deploy - agent sat there for 9.4 minutes until manual kill. added timeout wrapper at 45s and it works fine now but honestly this should be built into mcp by default. what timeout value are you using in your wrapper?