Context-aware AI coding

Claude Code doesn't know
your codebase.
Cortex does.

One command gives Claude Code full project context — git history, risk scores, PR decisions, security notes. Generated from your actual code, not written by hand.

$

The problem

Every session
starts from zero

  • Which parts of the codebase break most often
  • Why the auth module was rewritten three times
  • What that PR comment meant by "don't touch the cache"
  • Which files have no tests

You end up explaining context that should just exist.

You
Refactor the payment module. It's causing bugs.
Claude Code
I'll take a look. What kind of bugs are you seeing? Also, is there anything I should know about this module's history or architecture before I start?
You
It was rewritten twice, has no tests, and there's a known race condition in checkout. Oh, and there's a security note from a PR in January...

The solution

Context that travels with your code

Cortex analyzes your repository once and writes .claude/docs/ — a structured knowledge base Claude Code reads automatically.

It extracts what matters: bug patterns from git history, architectural decisions from PR discussions, risk scores, security issues.

Not hand-written documentation. Actual project history.

What it does

Everything Claude Code needs to know

🔍

Git History Mining

Scans every commit that touched a file. Bug fixes, refactors, regressions — all mapped to the files that caused them.

⚠️

Risk Scoring

Every file gets a score based on bug history, test coverage, churn rate, and dependency depth. HIGH / MEDIUM / LOW. Claude Code sees it before touching anything.

💬

PR Decisions

Mines GitHub PR comments for architectural decisions. "We chose Redis because…", "Don't cache this endpoint — caused prod outage". Knowledge that lives nowhere in the code.

🔒

Security Audit

Per-file SAST analysis. Hardcoded secrets in git history, SQL injection patterns, vulnerable dependencies. Results in context, not a separate report.

⏱️

Freshness Tracking

Every context file shows when it was generated and how many commits happened since. ⚡ Fresh / ⚠️ Stale / ❌ Outdated. You know if the context can be trusted.

🔌

MCP Integration

Run cortex mcp and add one line to .claude/settings.json. Context is injected automatically — no manual file reading required.

See it in action

Run one command

~/myproject
$ cortex analyze Analyzing 47 files... ✓ 47 files analyzed in 8.3s ✓ Languages: dart 31  typescript 16 ⚠ 4 files without tests ⚠ 2 security issues found — run cortex security ─────────────── Next Steps ─────────────── Quick setup wizard:   cortex setup Or configure MCP manually:   {"mcpServers": {"cortex": {"command": "cortex", "args": ["mcp"]}}} Commit:   git add .claude/ && git commit -m 'add cortex context'

Get started

Three steps

1

Install

One command. No configuration needed.

curl -fsSL https://raw.githubusercontent.com/kovalenko-tech/cortex/main/install.sh | bash
2

Analyze

Run in your project directory. Takes 30–120 seconds. Writes .claude/docs/ with context for every source file.

cortex analyze
3

Commit

Context travels with your repo. Everyone on the team benefits immediately.

git add .claude/
git commit -m "add cortex context"

Ready?

Give Claude Code the context it needs

Stop re-explaining your codebase every session. Run Cortex once and commit the result.

$ curl -fsSL https://raw.githubusercontent.com/kovalenko-tech/cortex/main/install.sh | bash