CodeGraph Live

Always-on semantic code intelligence for Claude Code

Fork of colbymchenry/codegraph — all the original features plus a persistent filesystem watcher daemon that keeps your graph fresh 24/7, even when Claude Code is closed.

What this fork adds

The upstream codegraph syncs your code graph in two ways:

After Claude Code edits a file (via a PostToolUse hook)
When you manually run codegraph sync

This means edits you make yourself — in VS Code, vim, or anywhere outside Claude — pile up undetected until you open Claude Code again.

codegraph-live adds a systemd user service that runs a background daemon. The daemon uses @parcel/watcher (native inotify on Linux) to watch all your codegraph-initialized projects simultaneously. Every file save is detected within 1 second, debounced, and synced — regardless of whether Claude Code is open.

8am  - Boot → daemon starts automatically, watching all your projects
9am  - You edit files in VS Code (Claude Code closed)
         → daemon catches every save in real time
11am - You open Claude Code
         → graph is already up to date, zero sync needed

Differences from upstream

Feature	upstream codegraph	codegraph-live
Sync on Claude edits	Yes (PostToolUse hook)	Yes
Sync on your own edits	No (manual `sync` only)	Yes — always automatic
Graph fresh when Claude is closed	No	Yes
Filesystem watcher	No	@parcel/watcher (inotify)
Systemd user service	No	Yes — starts at login
Auto-discover new projects	No	Yes — scans $HOME every 30s
CLI binary	`codegraph`	`codegraph-live`
All original features	—	Fully preserved

Installation

1. Install

Via npm (recommended):

npm install -g @evannsmc/codegraph-live

Or via npx (no global install):

npx @evannsmc/codegraph-live

From source:

git clone git@github.com:evannsmc/codegraph-live.git
cd codegraph-live
npm install
npm run build
npm install -g .

2. Register the systemd service

codegraph-live daemon install-service

This writes ~/.config/systemd/user/codegraph-live.service and enables it immediately. The daemon will start automatically at every login from now on.

3. Run the global installer

codegraph-live install

This is a one-time setup that writes to your global ~/.claude/ directory:

~/.claude/CLAUDE.md — CodeGraph usage instructions (applied in every project)
~/.claude.json — MCP server registration
~/.claude/settings.json — tool permissions and auto-sync hooks

You only need to run this once per machine. It will also offer to initialize the current directory as a project.

4. Initialize each project

Inside each project you want to track:

cd ~/your-project
codegraph-live init

This writes project-local config (.claude.json, .claude/settings.json) and indexes the codebase. The daemon picks up the new project within 30 seconds — no restart needed.

Daemon management

codegraph-live daemon status           # Show PID + list of watched projects
codegraph-live daemon start            # Start daemon manually (if not using systemd)
codegraph-live daemon stop             # Stop the daemon
codegraph-live daemon restart          # Restart

codegraph-live daemon install-service  # Register + enable systemd user service
codegraph-live daemon uninstall-service # Remove the service

Logs: ~/.local/share/codegraph-live/daemon.log PID file: ~/.local/share/codegraph-live/daemon.pid Service file: ~/.config/systemd/user/codegraph-live.service

How the daemon works

On startup — scans $HOME recursively for directories containing .codegraph/codegraph.db (the marker that a project has been initialized)
Per project — opens a @parcel/watcher subscription on the project root. Uses Linux inotify under the hood: near-zero CPU when idle
On file change — debounces 1000ms (batches rapid saves from formatters), then calls CodeGraph.open(root).sync() — the same incremental sync the upstream CLI uses
Auto-unwatch — if .codegraph/ is deleted (project uninstalled), the daemon removes it from the watch set silently
Auto-discover — every 30 seconds, rescans $HOME for newly initialized projects. New projects are picked up without restarting the daemon

The daemon coexists safely with the MCP server and any CLI sync calls via codegraph's existing Mutex + FileLock mechanism — concurrent syncs are serialized, not duplicated.

Ignored paths: node_modules, .git, dist, build, __pycache__, target, .cargo, *.pyc, *.o, *.so

Full CLI reference

All upstream commands are available under the codegraph-live binary.

Setup

codegraph-live install              # One-time global setup: writes ~/.claude/CLAUDE.md + MCP config
                                    # Run once per machine; also offered on bare `codegraph-live` invocation
codegraph-live init [path]          # Per-project: indexes code + writes local .claude.json/hooks
codegraph-live uninit [path]        # Remove CodeGraph from a project (deletes .codegraph/)

Indexing & sync

codegraph-live reindex [path]       # Fully rebuild the graph from scratch (all files)
codegraph-live sync [path]          # Incremental sync — re-index only files that changed
codegraph-live status [path]        # Show index stats (file count, nodes, DB size, last sync)

With the daemon running, you rarely need either manually — the daemon syncs on every file save automatically.

Querying

codegraph-live query <search>       # Search for symbols by name (functions, classes, types)
codegraph-live files [path]         # Show project file structure as known to the graph
codegraph-live explain <task>       # Build rich context for a task — finds relevant symbols,
                                    #   call graphs, and code semantically (outputs markdown)
codegraph-live test-changes [files...]  # Find test files that cover the given source files
                                        #   (reverse graph traversal — useful in CI/pre-commit)

test-changes example:

# After editing a planning module, find which tests cover it:
git diff --name-only HEAD~1 | xargs codegraph-live test-changes

MCP server & internal commands

codegraph-live serve --mcp          # [Internal] Claude Code starts this automatically — you never run it directly
codegraph-live mark-dirty [path]    # [Internal] Called by Claude Code's PostToolUse hook
codegraph-live sync-if-dirty [path] # [Internal] Called by Claude Code's Stop hook

MCP tools available inside Claude Code sessions:

Tool	Purpose
`codegraph_explore`	Deep exploration — comprehensive context for a topic in one call
`codegraph_context`	Quick context for a task
`codegraph_search`	Find symbols by name (functions, classes, types)
`codegraph_callers`	Find what calls a function
`codegraph_callees`	Find what a function calls
`codegraph_impact`	Find what's affected by changing a symbol
`codegraph_node`	Get details + source for a specific symbol

Daemon (new in this fork)

codegraph-live daemon start             # Start daemon manually (detached background process)
codegraph-live daemon stop              # Stop the running daemon
codegraph-live daemon restart           # Restart the daemon
codegraph-live daemon status            # Show PID, running state, and list of watched projects

codegraph-live daemon install-service   # Write + enable ~/.config/systemd/user/codegraph-live.service
codegraph-live daemon uninstall-service # Disable and remove the systemd service

Supported languages

TypeScript, JavaScript, TSX, JSX, Python, Go, Rust, Java, C, C++, C#, PHP, Ruby, Swift, Kotlin, Dart, Liquid, Pascal (19 total)

Staying up to date with upstream

git remote add upstream https://github.com/colbymchenry/codegraph.git
git fetch upstream
git merge upstream/main
npm run build
npm install -g .
codegraph-live daemon restart

Credits

colbymchenry/codegraph — the upstream project this fork is based on. All core graph indexing, MCP tools, semantic search, and Claude Code integration come from there.
sdsrss/code-graph-mcp — inspiration for the always-on daemon architecture. That project's persistent filesystem watcher (implemented in Rust) demonstrated that keeping the graph live between sessions is both practical and valuable. The daemon in this fork applies the same idea to the codegraph TypeScript ecosystem.

License

MIT — same as upstream. Fork by evannsmc.

Name		Name	Last commit message	Last commit date
Latest commit History 241 Commits
__tests__		__tests__
architecture		architecture
docs		docs
scripts		scripts
src		src
.gitignore		.gitignore
CLAUDE.md		CLAUDE.md
DELPHI-SUPPORT.md		DELPHI-SUPPORT.md
LICENSE		LICENSE
README.md		README.md
package-lock.json		package-lock.json
package.json		package.json
publish.js		publish.js
tsconfig.json		tsconfig.json
vitest.config.ts		vitest.config.ts

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

CodeGraph Live

Always-on semantic code intelligence for Claude Code

What this fork adds

Differences from upstream

Installation

1. Install

2. Register the systemd service

3. Run the global installer

4. Initialize each project

Daemon management

How the daemon works

Full CLI reference

Setup

Indexing & sync

Querying

MCP server & internal commands

Daemon (new in this fork)

Supported languages

Staying up to date with upstream

Credits

License

About

Uh oh!

Releases

Packages

Uh oh!

Contributors

Uh oh!

Languages

Folders and files

Latest commit

History

Repository files navigation

CodeGraph Live

Always-on semantic code intelligence for Claude Code

What this fork adds

Differences from upstream

Installation

1. Install

2. Register the systemd service

3. Run the global installer

4. Initialize each project

Daemon management

How the daemon works

Full CLI reference

Setup

Indexing & sync

Querying

MCP server & internal commands

Daemon (new in this fork)

Supported languages

Staying up to date with upstream

Credits

License

About

Topics

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages