Install Claude-Mem in Claude Code
Install Claude-Mem safely from its marketplace, inspect its storage path, and test the first context recovery.
- Skill Road
- Install Claude-Mem in Claude Code
Published on 09.09.2026
Claude-Mem is installed in Claude Code through its operator’s marketplace. The repository now calls the product Grok Mem, but the marketplace and package name remain claude-mem. This guide uses the installable name and starts with a non-sensitive test project.
Check prerequisites
Per the official README, the supported path needs a current Claude Code version with plugin support and Node.js 20 or later. Check Node with node --version. The installer can add Bun and uv when needed; restrictive enterprise networks should clear that with IT first.
Add the marketplace and install the plugin
Open Claude Code in the target project and add the official marketplace:
/plugin marketplace add thedotmack/claude-mem
Then install the plugin:
/plugin install claude-mem
Restart Claude Code afterward. The official README also documents npx claude-mem install; that route configures the plugin, hooks, and worker together. npm install -g claude-mem is insufficient because, per the provider, it installs only the SDK or library.
Choose the storage and model path deliberately
The installer can configure a hosted observer, your own model access, or an Anthropic-plan route. The documentation also describes local SQLite and Chroma data plus optional cloud sync. Before using a production project, establish where data lives, whether the local viewer is reachable, and whether an external service participates. Cloud sync, a hosted observer, and external models each add their own privacy and retention terms.
Test the first run safely
Begin with a harmless, bounded task, such as creating a test note or inspecting a non-sensitive file. End or restart Claude Code, then ask a focused question about the work just completed. That checks whether context is captured and found again without bringing production secrets into the test. Afterwards, inspect storage placement and viewer access.
Exclude sensitive content
Claude-Mem observes prompts and tool use. Mark sensitive sections with <private> as documented by the provider so they are not stored. Add the usual controls: do not put keys in prompts or logs, use least-privilege tools, protect local database files, and keep secret scanning enabled in the repository. The tag does not replace a review of whether the selected external model or hosting route is authorized for the data.
Frequently asked questions
Is Claude-Mem a valid Claude Code plugin?
Yes. The official `.claude-plugin/marketplace.json` in the thedotmack/claude-mem repository lists `claude-mem` as a plugin sourced from `./plugin`.
Why not simply install it globally with npm?
Per the official README, `npm install -g claude-mem` installs only the SDK or library. Use `/plugin install claude-mem` after adding the marketplace, or use `npx claude-mem install`, for hooks and the worker.
Are confidential prompts stored?
The plugin observes prompts and tool use. The source documents `<private>` for excluding content; storage location, participating model providers, and cloud sync still need deliberate review.