zcodegraph MCP returns malformed database after reindex

[jununfly]

Problem

In jununfly/ZAgenticLoop, zcodegraph_explore failed with:

Error: Tool execution failed: database disk image is malformed

I then ran:

zcodegraph index

The CLI completed successfully:

Indexed 49 files
702 nodes, 1,213 edges
Indexed with rust-hybrid
Fallback health: degraded
27 TypeScript fallback files

But a later MCP zcodegraph_explore call still returned the same malformed database error.

Expected

After a successful zcodegraph index, MCP queries should use the rebuilt usable index, or surface a clear remediation step if the MCP server is still holding a stale/corrupt SQLite handle.

Actual

MCP continues to report database disk image is malformed even after CLI reindex succeeds.

Context

• Project: jununfly/ZAgenticLoop
• Command run from repo root: zcodegraph index
• Follow-up MCP call: zcodegraph_explore
• CLI index result was usable but fallback-degraded.

Possible cause

The MCP process may keep an old/corrupt DB connection or cache after the CLI rebuilds the index, or the reindex may not replace the same database file used by the MCP server.

Suggested behavior

• Detect and recover from malformed SQLite DB by reopening/rebuilding.
• If recovery is unsafe, return a specific remediation such as restarting MCP or running a particular repair command.
• Consider verifying that CLI and MCP resolve the same .zcodegraph path.