tana_sync MCP tool crashes supertag-mcp process after successful completion

Version: 2.5.10

Description

Calling the tana_sync MCP tool causes the supertag-mcp process to crash immediately after the tool completes successfully. The tool itself runs to completion (the sync succeeds and is logged as successful), but the process then emits a malformed response that terminates the MCP connection. All subsequent tool calls on that session fail.

Steps to reproduce

1. Connect to supertag-mcp via an MCP client (tested via agentgateway proxying stdio)
2. Call the tana_sync tool
3. Tool completes and logs success
4. supertag-mcp process exits/crashes

Observed behaviour

The agentgateway transport layer logs:

Error reading from stream: serde error expected value at line 1 column 1

All subsequent calls on the session return upstream closed on receive. The supertag-mcp process is dead.

Expected behaviour

tana_sync completes and the process remains alive and responsive.

Notes

• The sync itself succeeds — data is indexed correctly. This is purely a post-completion crash in the MCP response serialisation.
• supertag-mcp's own logs confirm Tool executed successfully before the crash — the error is in the final response write, not the sync logic itself.
• The serde error (expected value at line 1 column 1) indicates supertag-mcp is closing stdout or exiting without writing a valid JSON response after the tool completes.
• This is reproducible through agentgateway (stdio transport) but the root cause is in supertag-mcp, not the transport layer — stdio is standard MCP transport.
• Workaround: kill the process and reconnect — the new session works normally.

Start at the tana_sync MCP tool entry point and trace the response written after the successful sync, using the stdio transport and the serde error expected value at line 1 column 1 failure as the signal. Reproduce through an MCP client or agentgateway, then verify that a valid response is emitted and the process remains available for a second tool call.