ProjectCtx documentation

ProjectCtx gives AI tools durable context across conversations. It stores workspace-scoped memory behind a shared web, REST, and MCP service so you can inspect what your tools know and correct it when needed.

ProjectCtx is currently invite-only and the command-line agent is distributed on npm's pre-1.0 alpha channel.

Start In Five Minutes#

1. Open projectctx.com, activate the sign-in screen, and select Have an invite code?
2. Enter your email address and the invite code supplied by your workspace administrator. Follow the secure email link to choose a password.
3. Install the ProjectCtx agent:

npm install -g @projectctx/agent@alpha

4. Preview and apply the supported-client configuration:

projectctx connect --dry-run
projectctx connect
projectctx doctor

5. Restart each configured AI client. The first hosted ProjectCtx tool call opens that client's OAuth authorization flow.

For the complete walkthrough, continue to Getting started.

What You Can Connect#

The agent detects and configures these local clients:

• Claude Code
• Cursor
• Codex
• Hermes / Jarvis

ChatGPT, claude.ai, and Claude Desktop use account-side connector settings. projectctx connect prints the hosted MCP URL and the managed instruction text to add in those clients.

See Connect AI tools for the complete setup model.

What Runs Locally#

Repository scanning is read-only and runs only when you invoke it. ProjectCtx does not install a repository watcher, continuous indexer, or file-sync daemon.

Supported local clients can also install lifecycle hooks for automatic memory. Those hooks stage a small set of durable outcomes instead of storing complete transcripts.

Durable repository memory uses a separate preview and apply workflow, so you can inspect proposed records and relationships before they are written.

Find A Specific Answer#

• CLI reference
• Security
• Troubleshooting

Maintainer release and deployment procedures are intentionally excluded from the public documentation.