Does it sync automatically?

[hohowt]

Will Codegraph sync automatically after I update the code in the project?
Or do I have to manually run codegraph sync every time I update the code?

[hubianluanma]

Based on my usage, it does not automatically update by default. Every time I make a code change, I will execute the codegraph index. However, I think a better approach is to listen for file changes and then automatically update it

[krzkz94]

Would be nice if we could have a --watch flag to have it running in the background as it's making changes, would be very helpful, especially if it can be debounced, like --watch --debounce 1500

[hubianluanma]

Indeed, AI suggested that I use the watchexec tool to listen for files of a specified type, like this:
watchexec
-r
--debounce 800ms
--watch src
--exts ts,tsx,js,jsx,json
--ignore node_modules
--ignore dist
--ignore build
--ignore .git
'codegraph index'

[colbymchenry]

It uses a FileWatcher and auto syncs on every file change. It's constantly re-indexing and syncing. Closing this issue.

[colbymchenry]

Confirmed — codegraph auto-syncs on every file change, no manual codegraph sync/index needed. The MCP server runs a built-in FileWatcher that incrementally re-indexes and re-resolves edges on each save.

Validated end-to-end on the released 0.9.4 npm build across all three OSes — install → start the MCP server → edit a file → the new symbols and their calls/imports edges appear in callees/query within ~3s, with existing edges intact:

• ✅ macOS — FSEvents
• ✅ Linux — inotify (tested in a clean node:22 Docker container)
• ✅ Windows 11 — ReadDirectoryChangesW

One practical note: the watcher lives inside the long-running MCP server process, which loads its code at startup. After upgrading codegraph, restart your editor / MCP client so the new version (and watcher) is picked up.

[colbymchenry]

Follow-up from validating this — I found a gap in how the auto-sync detects changes, and fixed it.

Incremental sync (both the file watcher and the git hooks call codegraph sync) was detecting changes via git status, which only sees uncommitted working-tree changes. So after a git pull / checkout / merge / rebase — where the working tree ends up clean — sync reconciled nothing, and the index silently went stale until a full codegraph index -f. It also meant non-git projects leaned on a slower full-rescan path.

Change detection is now filesystem-based and git-independent: a cheap (size, mtime) stat pre-filter skips unchanged files, then a content-hash check confirms the rest. The MCP server also now runs a catch-up sync on connect, so anything changed while it wasn't running (e.g. you pulled from the terminal, then opened your editor) is reconciled right away.

Validated end-to-end — each case was stale before, reconciled after; no-op syncs stay incremental and existing edges are preserved:

• ✅ committed change with a clean working tree (pull-like)
• ✅ branch switch (symbols added on the branch, removed on switch-back)
• ✅ plain edit in a non-git project
• ✅ catch-up sync on MCP connect (Caught up 1 file(s) changed since last run)

Fix is implemented and validated; it'll ship in an upcoming release.

[hubianluanma]

It'd be great if codegraph status could show the latest update date and time. That way, we can make sure the automatic sync is actually working.