claude-code - 💡(How to fix) Fix linear-server MCP tool calls intermittently fail with user-rejection error during long sessions (hang / block / disconnect — root cause unknown) [2 comments, 3 participants]

During long Claude Code sessions using the linear-server MCP, tool calls (list_issues, save_comment, save_issue) intermittently return an error shape identical to a user-initiated rejection: The user did not reject the tool call. From the agent side we cannot distinguish whether the MCP server hung, timed out, blocked, or disconnected — the error text is the same.

• After ~30-60 minutes of active use, tool calls begin failing intermittently with the user-rejection error shape.

1. A distinct error signal for transport-level MCP failures vs genuine user rejection, so the agent doesn't misattribute causality.

TL;DR

Implement a retry mechanism with a distinct error handling for transport-level MCP failures to differentiate them from genuine user rejections.

Guidance

• Investigate the linear-server MCP connection management to identify potential causes of hang, block, or disconnect patterns during long sessions.
• Consider implementing a heartbeat or keep-alive mechanism to maintain the connection and prevent timeouts.
• Develop a custom error handling mechanism to catch and retry tool calls that fail with the user-rejection error shape, allowing the agent to distinguish between genuine user rejections and transport-level failures.
• Analyze the conversation log to identify patterns or correlations between the false-positive user-rejection signals and other system events.

Example

# Pseudo-code example of a retry mechanism with custom error handling
def tool_call_retry(tool_call):
    max_retries = 3
    for attempt in range(max_retries):
        try:
            # Perform the tool call
            response = perform_tool_call(tool_call)
            if response.error_shape == "user_rejection":
                # Custom error handling for transport-level failures
                if is_transport_failure(response):
                    # Retry the tool call
                    continue
                else:
                    # Genuine user rejection, stop and wait for user input
                    break
            else:
                # Successful tool call, return the response
                return response
        except Exception as e:
            # Handle other exceptions
            print(f"Error: {e}")
    # Max retries exceeded, raise an error
    raise Exception("Max retries exceeded")

def is_transport_failure(response):
    # Custom logic to determine if the error is a transport-level failure
    # This could involve analyzing the error message, response headers, or other factors
    pass

Notes

The provided example is a pseudo-code illustration and may require modifications to fit the actual implementation. The is_transport_failure function is a placeholder for custom logic to determine if the error is a transport-level failure.

Recommendation

Apply a workaround by implementing a retry mechanism with custom error handling, as this will allow the agent to distinguish between genuine user rejections and transport-level failures, reducing the impact of false-positive user-rejection signals.