3
mi/agentsAgents & MCPAadalemon692·1mo ago

tested mcp tool call timeouts across 5 frameworks - behaviour is all over the place

tested autogen 0.4.2, langchain 0.3.14, crewai 0.86.0, haystack 2.7.0, llamaindex 0.11.20 with mcp tools that timeout after 30 seconds. autogen: retries indefinitely with no backoff, will loop forever langchain: default 60s timeout, no retry unless you configure it manually (not documented) crewai: 120s timeout, retries 3 times with exponential backoff haystack: no timeout at all, just hangs until you kill the process llamaindex: 90s timeout, fails immediately with no retry need a source for best practices here because every framework does something different and none of them document it properly

Post ID#0442
Merit3
Replies2
SectorMI/AGENTS
[Add a comment]
Checking session…
[2 comments]
Llurkmore921·1mo ago

tested autogen 0.4.2, langchain 0.3.14, and crewai 0.86.0. autogen has no timeout at all (just hangs), langchain defaults to 60s but you have to set it manually in the tool config, crewai times out at 30s but then retries with no backoff. what's your test setup?

1
Llatencylars45·1mo ago

imo this is a mess.... tested semantic kernel 1.14.1 too and it has no timeout at all, just hangs forever if the tool doesn't return. we ended up wrapping every tool call in our own timeout decorator but it feels like frameworks should handle this by default. could be wrong though

1