CLI to initialize and manage Cortex Context (Knowledge Graph) tooling in any VS Code workspace.
Installs:
- Skills — VS Code Copilot Agent skills for Cortex (
cortex-research,cortex-dimensions,cortex-ingest,generate-spec) - MCP Server — TypeScript stdio server exposing Cortex tools to GitHub Copilot (bundled — no Python needed)
- VS Code Config — Injects
cortexentry into.vscode/mcp.json
- Node.js 18+
- A running Cortex server
# Install globally
npm install -g @cortex-context/cli
# Or run directly without installing
npx @cortex-context/cli init# Initialize Cortex Context in the current workspace
cortex-context init
# Initialize with specific options (non-interactive)
cortex-context init --url http://localhost:8082 --token <your-token>
# Initialize in a specific workspace folder
cortex-context init --workspace /path/to/your/workspace
# Sync Skills/MCP to latest bundled version
cortex-context sync
# Preview what sync would change
cortex-context sync --dry-run
# Validate installation and connectivity
cortex-context doctor- Prompts for Cortex server URL and optional API token
- Tests connectivity to the Cortex server
- Copies Skills into
{workspace}/.github/skills/ - Injects
cortexentry into{workspace}/.vscode/mcp.json(usesnpx @cortex-context/cli mcp-serve) - Saves configuration to
~/.cortex-context/config.json
# Reload VS Code (Cmd/Ctrl+Shift+P → "Reload Window")
# Then run doctor to verify:
cortex-context doctor| Command | Description |
|---|---|
cortex-context init |
Full interactive setup (Skills, MCP, hooks, rules) |
cortex-context update |
Update Skills + MCP files to latest bundled version |
cortex-context update --dry-run |
Preview update changes without writing files |
cortex-context sync |
Ingest latest git diff into the Cortex Knowledge Graph |
cortex-context sync --dry-run |
Show diff without sending to Cortex API |
cortex-context doctor |
Validate installation + test Cortex connectivity |
cortex-context uninstall |
Remove Cortex Context from the workspace |
| Command | Description |
|---|---|
cortex-context server |
Install Docker (if needed) and start Neo4j + Cortex API locally |
cortex-context server --embeddings |
Start with embeddings image (sentence-transformers, CPU, ~1.5 GB) |
cortex-context server --embeddings-gpu |
Start with embeddings GPU image (PyTorch CUDA, ~2 GB — requires nvidia) |
cortex-context upgrade |
Pull latest Cortex backend image and restart the local stack |
cortex-context upgrade --embeddings |
Upgrade using the embeddings image variant |
cortex-context upgrade --embeddings-gpu |
Upgrade using the embeddings GPU image variant (PyTorch CUDA, ~2 GB — requires nvidia) |
cortex-context upgrade --pull-only |
Pull latest image without recreating the container |
cortex-context downgrade |
Switch to the base image variant (lightest, no embeddings) |
cortex-context downgrade --embeddings |
Switch to the CPU embeddings variant (lighter than GPU) |
| Command | Description |
|---|---|
cortex-context token |
Show current API token (masked) |
cortex-context token --show |
Reveal the full token |
cortex-context token --rotate |
Generate a new random token, save it, and restart stack |
cortex-context token --set <val> |
Set a specific token value (empty string clears auth) |
| Command | Description |
|---|---|
cortex-context reset |
Wipe all nodes and relationships from the Knowledge Graph |
cortex-context mcp-serve |
Start the MCP server over stdio (used internally by VS Code/Cursor) |
| Flag | Description |
|---|---|
--url <url> |
Cortex server URL (skips prompt) |
--token <tok> |
API token (skips prompt) |
--workspace |
Target workspace path (default: cwd) |
--local |
Deploy a local Cortex stack with Docker (mutually exclusive with --url) |
--skip-mcp |
Don't configure MCP server |
--skip-skills |
Don't install Skills |
--skip-hooks |
Don't install git hooks |
--skip-rules |
Don't inject Copilot/Cursor rules |
--force |
Overwrite existing files without prompting |
Contributions are welcome! Please read CONTRIBUTING.md before opening a pull request.
Found a bug? Open a bug report.
Have an idea? Open a feature request.
MIT