linear-server MCP tool calls intermittently fail with user-rejection error during long sessions (hang / block / disconnect — root cause unknown)

Observed symptom

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 doesn't want to proceed with this tool use.
The tool use was rejected (eg. if it was a file edit,
the new_string was NOT written to the file).
STOP what you are doing and wait for the user to tell
you how to proceed."

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.

Reproduction

• Long Claude Code session with frequent linear-server tool invocations.
• After ~30-60 minutes of active use, tool calls begin failing intermittently with the user-rejection error shape.
• /mcp reconnect restores the tool set; after reconnect the same tool calls succeed.

Impact

• False-positive user-rejection signal: the agent treats the failure as user intervention and stops / pivots incorrectly. The conversation log records a "user rejection" that the user never issued. Agents using the log for behavior-tuning learn the wrong lesson.
• Mid-turn interruption during autonomous or peer-review loops, requiring a manual /mcp reconnect to resume.

Workaround

/mcp reconnect when the failure shape appears.

Environment

• Claude Code CLI (version TBD)
• linear-server MCP (official Linear hosted server)
• Linux (WSL2), long-running interactive session

Ask

1. A distinct error signal for transport-level MCP failures vs genuine user rejection, so the agent doesn't misattribute causality.
2. Root-cause investigation of the hang / block / disconnect pattern on long sessions.