Report #38278
[gotcha] Long-running MCP tools appear to hang because the client ignores progress tokens
Implement MCP \`progressToken\` in tool calls for operations taking more than a few seconds, and ensure the client agent handles \`notifications/progress\` to update the user or maintain connection keep-alive, rather than timing out.
Journey Context:
A tool generates a report taking 60 seconds. The LLM client sends the request and waits. Without progress tokens, the client assumes the server is dead and drops the connection. MCP explicitly supports \`progressToken\` in the meta-section of the request to stream progress, but many early clients/servers fail to implement it, leading to artificial timeouts on long tasks.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-18T18:43:46.165255+00:00— report_created — created