mcp 0.5.0 - tool timeout configuration is ignored, tested with slow filesystem operations
1. tested mcp 0.5.0-patch2 with a custom filesystem tool that scans large directories 2. set tool timeout to 5000ms in the tool schema 3. agent calls the tool on a directory with 18k files, tool takes 34 seconds to respond 4. timeout config is completely ignored - agent just waits indefinitely the tool schema has `"timeout": 5000` but the agent doesn't enforce it. we had to patch manual timeout handling at the wrapper layer with a promise race. is this a known limitation or are we missing a config option somewhere?
tested this yesterday with filesystem tool reading large directories (12k+ files). set timeout to 5000ms in tool schema but operations that take 18+ seconds just keep running until they complete. is there config option to enforce timeout or does schema constraint do nothing?
same issue with s3 tool doing large uploads - set timeout to 60s, uploads that take 4+ minutes just keep running. config is completely ignored and there's no way to kill it except process termination
That's brutal - 4 minute uploads with no timeout enforcement is a production nightmare. Does the timeout config at least get logged somewhere so you can see it's being ignored, or is it just silently not working?
ok so the timeout config gets logged in the mcp server logs but it's completely ignored at runtime. you can see "timeout: 60000" in the startup logs and then watch operations run for 4+ minutes anyway. zero enforcement
tested with filesystem tool doing 8gb file reads. set timeout to 30s in config, tool ran for 340 seconds before i killed it. config is completely ignored