v0.0.1 CLI · v1 Graph + Vector RAG · MCP-native

Product Knowledge Graph for AI Agents

Cortex Context

Give GitHub Copilot and any AI coding agent deep, structured understanding of your product — specs, services, decisions and workflows — in a queryable graph.

Install on npm ↗ See how it works

npx @cortex-context/cli@latest init

Terminal

          $ npx
          @cortex-context/cli@latest
          init


            ◆ Phase 1 Environment
 ✓ .env
          CORTEX_URL=http://localhost:8082

            ✓ cortex.config.yaml
          ejected

            ✓ docker compose up
          Neo4j + Cortex API


            ◆ Phase 2 MCP + Rules
 ✓ copilot-instructions
          Cortex block injected

            ✓ .vscode/mcp.json
          configured


            ◆ Phase 3 Git hook
 ✓ .git/hooks/post-commit
          auto-sync installed


            Done in 6.1s → cortex-context doctor
        

I.S.I.R Ontology

Four pillars. Complete context.

Every piece of product knowledge maps to one of four universal concepts, giving AI agents a structured map of your entire codebase.

🎯

Pillar 1

Intent

The "Why"

Specs, requirements and user stories that explain the purpose behind every feature.

:Spec:UserStory:Epic

🏗️

Pillar 2

System

The "How"

Services, ADRs and APIs that form the structural skeleton of your product.

:Service:ADR:API

⚙️

Pillar 3

Implementation

The "What"

Modules, workflows and components — the concrete building blocks that make it run.

:Workflow:Module:Component

📡

Pillar 4

Runtime

The "Real world"

Alerts, deployments and incidents that ground AI responses in what's actually happening.

:Alert:Deployment:Incident

CLI

One command. Full setup.

A TypeScript CLI that wires your workspace to the knowledge graph in three phases: local Docker stack, MCP server config, and a git hook for automatic sync on every commit.

$ npx @cortex-context/cli@latest init npm ↗

cortex-context init 3-phase setup: local Docker stack + MCP server (Copilot / Claude) + post-commit git hook. Run once per workspace.

cortex-context sync Reads git diff HEAD~1, auto-extracts spec_ref from the branch name, and ingests the diff into the graph. Called automatically by the hook.

cortex-context sync --dry-run Previews the diff and detected spec_ref without sending to the server. Useful for debugging the hook.

cortex-context update Updates Skills and the MCP Server bundle in your workspace to the latest published version.

cortex-context doctor Checks Cortex connectivity, validates the local install, and reports the hook and MCP server status.

⚡ Example output

$ cortex-context sync --repo . → spec_ref detected: spec-151 ← branch: feature/151-spec150-port ✓ Diff captured 32 insertions · 4 files ✓ Ingested in 134ms POST /api/v1/ingest (:Commit)-[:IMPLEMENTS]->(:Spec {id:"spec-151"}) $ cortex-context doctor ✓ Cortex URL http://10.11.12.200:8082
✓ Neo4j bolt://localhost:7687
✓ Nodes in graph 150 (122 specs · 8 svc · 15 wf)
✓ Embedder local (sentence-transformers)
✓ MCP server stdio — GitHub Copilot / Claude
✓ Git hook post-commit installed

Features

Built for the AI-native team

🕸️

Graph-first exploration

FTS hits trigger 1–2 hop expansions — your agent sees a spec plus everything it impacts: services, decisions, workflows.

🔬

Hybrid RAG

Combine full-text precision with semantic similarity. Built-in local embeddings or OpenAI for natural-language queries.

🔌

MCP-native

Ships with a Model Context Protocol server. GitHub Copilot, Claude Desktop and any MCP agent connects out of the box.

🧩

Plugin dimensions

Add GitHub Issues, Confluence or Jira as a single YAML + Python file. No core code changes required.

🔒

Fully local

Everything runs on your machine or private server. Your source code never leaves your infrastructure.

📈

Impact analysis

Ask "which services are affected by spec-070?" and get a precise answer via 1–2 hop graph traversal — no grep, no guessing.

🔗

Branch-aware spec tracking

sync reads the branch name, extracts the spec ID from the feature/NNN-slug convention, and writes a (:Commit)-[:IMPLEMENTS]->(:Spec) edge automatically.

⏱️

Bitemporal versioning

Every node carries ingested_at and valid_from — know when the data was added and when the spec was written.

Integrations

Works with your stack

Product-agnostic and open. Plug it into the tools your team already uses.

🤖 GitHub Copilot

🧠 Claude Desktop

⚡ VS Code

🕸️ Neo4j

🐍 FastAPI

🐳 Docker

🐙 GitHub

📄 Markdown / YAML

🔷 OpenAI Embeddings

📦 npm

🔍 query_product_context

🔍 get_spec_context

🔍 get_service_context

📊 Any MCP client

Quick Start

Up in 5 minutes

Three commands to a working knowledge graph.

1Install the CLI

npx @cortex-context/cli@latest init # or: npm install -g @cortex-context/cli

2Start the graph

docker compose up -d # Starts Neo4j + Cortex API

3Init & sync

cortex-context init --url http://localhost:8082 # ✅ MCP configured + git hook installed cortex-context sync --repo . # ✅ Graph updated — Copilot has context