Release: v0.17.0
v0.17.0
Incremental index synchronization
Reconcile an existing index with disk instead of re-ingesting everything: new and changed files are re-ingested, byte-identical files are skipped, and entries whose source file is gone are removed.
npx mcp-local-rag sync [path] # omit the path to cover every configured rootOver MCP, sync_start returns a job id and sync_status is polled until state is no longer running. Both surfaces report upserted, skipped, empty, and pruned.
Upgrading
The first sync re-ingests everything. Chunks indexed before 0.17.0 carry no content hash, so they cannot be recognized as unchanged.
Pruning rules, scan warnings, the full status record, and the single-writer constraint are documented in the README and the skill reference.
No breaking changes, no new dependencies.