CKB — Code Knowledge Backend
Know your code. Change it safely. Ship with confidence.
CKB transforms your codebase into a queryable knowledge base. Ask questions, understand impact, find owners, detect dead code—all through CLI, API, or AI assistants.
Think of it as a senior engineer who knows every line of code, every decision, and every owner—available 24/7 to answer your questions.
Instant Answers to Hard Questions
| Question | Without CKB | With CKB |
|---|---|---|
| "What breaks if I change this?" | Grep and hope | Precise blast radius with risk score |
| "Who should review this PR?" | Guess from git blame | Data-driven reviewer suggestions |
| "Is this code still used?" | Delete and see what breaks | Confidence-scored dead code detection |
| "What tests should I run?" | Run everything (30 min) | Run affected tests only (2 min) |
| "How does this system work?" | Read code for hours | Query architecture instantly |
| "Who owns this code?" | Search CODEOWNERS manually | Ownership with drift detection |
| "Are there exposed secrets?" | Manual grep for patterns | Automated scanning with 26 patterns |
What You Can Do
🔍 Understand — Semantic search, call graphs, usage tracing, architecture maps
⚡ Analyze — Impact analysis, risk scoring, hotspot detection, coupling analysis
🛡️ Protect — Affected test detection, breaking change warnings, PR risk assessment
🔐 Secure — Secret detection, credential scanning, security-sensitive code identification
👥 Collaborate — Ownership lookup, reviewer suggestions, architectural decisions (ADRs)
📊 Improve — Dead code detection, tech debt tracking, documentation coverage
🚀 Compound Operations — Single-call tools (explore, understand, prepareChange) reduce AI tool calls by 60-70%
🔗 Integrate — CLI, HTTP API, MCP for AI tools, CI/CD pipelines, custom scripts
Try It Now
# See what's risky in your codebase
ckb hotspots --format=human
# Check what changed and what might break
ckb diff-summary --format=human
# Scan for exposed secrets
ckb audit --format=human
# Check architecture at a glance
ckb arch --format=human
# Check system status
ckb statusWorks Everywhere
| AI Assistants | CI/CD | Your Tools |
|---|---|---|
| Claude Code, Cursor, Windsurf, VS Code, Grok | GitHub Actions, GitLab CI | CLI, HTTP API, Scripts |
83% token reduction with smart presets—load only the tools you need.
# One command to connect to Claude Code
ckb setupBuilding your own tools? Use CKB as a backend via CLI, HTTP API, or MCP. See the Integration Guide for examples in Node.js, Python, Go, and shell scripts.
Learn More
| Resource | Description |
|---|---|
| 📖 Features Guide | Complete feature list with examples |
| 💬 Prompt Cookbook | Real prompts for real problems |
| 🔌 Integration Guide | Use CKB in your own tools and scripts |
| ⚡ Impact Analysis | Blast radius, affected tests, PR risk |
| 🔧 CI/CD Integration | GitHub Actions, GitLab CI templates |
Quick Start
Option 1: npm (Recommended)
# Install globally
npm install -g @tastehub/ckb
# Or run directly with npx (no install needed)
npx @tastehub/ckb initOption 2: Homebrew (macOS/Linux)
brew tap SimplyLiz/ckb
brew install ckbOption 3: Build from Source
git clone https://github.com/SimplyLiz/CodeMCP.git
cd CodeMCP
go build -o ckb ./cmd/ckbSetup
# 1. Initialize in your project
cd /path/to/your/project
ckb init # or: npx @tastehub/ckb init
# 2. Generate SCIP index (optional but recommended)
ckb index # auto-detects language and runs appropriate indexer
# 3. Connect to Claude Code
ckb setup # creates .mcp.json automatically
# Or manually:
claude mcp add --transport stdio ckb -- npx @tastehub/ckb mcpToken efficiency shown at startup:
CKB MCP Server v9.0.0
Active tools: 14 / 76 (18%)
Estimated context: ~1k tokens
Preset: coreNow Claude can answer questions like:
- "What calls the HandleRequest function?"
- "How is ProcessPayment reached from the API?"
- "What's the blast radius if I change UserService?"
- "Who owns the internal/api module?"
- "Is this legacy code still used?"
Why CKB?
| Without CKB | With CKB |
|---|---|
| AI greps for patterns | AI navigates semantically |
| "I found 47 matches for Handler" | "HandleRequest is called by 3 routes via CheckoutService" |
| Guessing at impact | Knowing the blast radius with risk scores |
| Reading entire files for context | Getting exactly what's relevant |
| "Who owns this?" → search CODEOWNERS | Instant ownership with reviewer suggestions |
| "Is this safe to change?" → hope | Hotspot trends + impact analysis |
Three Ways to Use It
| Interface | Best For |
|---|---|
| MCP | AI-assisted development — Claude, Cursor, Windsurf, VS Code, OpenCode, Grok |
| CLI | Quick lookups from terminal, scripting |
| HTTP API | IDE plugins, CI integration, custom tooling |
How Indexing Works
CKB uses SCIP indexes to understand your code. Think of it like a database that knows where every function is defined, who calls it, and how everything connects.
The Basics
# 1. Generate an index (auto-detects language)
ckb index
# 2. Check if your index is fresh
ckb statusWithout an index, CKB still works using tree-sitter parsing (basic mode), but with an index you get:
- Cross-file references ("who calls this function?")
- Precise impact analysis
- Call graph navigation
Language Support
Not all languages are equal. CKB classifies languages into quality tiers based on indexer maturity:
| Tier | Quality | Languages |
|---|---|---|
| Tier 1 | Full support, all features | Go |
| Tier 2 | Full support, minor edge cases | TypeScript, JavaScript, Python |
| Tier 3 | Basic support, call graph may be incomplete | Rust, Java, Kotlin, C++, Ruby, Dart |
| Tier 4 | Experimental | C#, PHP |
Key limitations:
- Incremental indexing is Go-only. Other languages require full reindex.
- TypeScript monorepos may need
--infer-tsconfigflag - C/C++ requires
compile_commands.json - Python works best with activated virtual environment
Run ckb doctor --tier standard to check if your language tools are properly installed.
See Language Support for indexer installation, known issues, and the full feature matrix.
Keeping Your Index Fresh
Your index becomes stale when you make commits. CKB offers several ways to stay current:
| Method | Command | When to Use |
|---|---|---|
| Manual | ckb index | One-off updates, scripts |
| Watch mode | ckb index --watch | Auto-refresh during development |
| MCP watch | ckb mcp --watch | Auto-refresh in AI sessions |
| CI webhook | POST /api/v1/refresh | Trigger from CI/CD |
Quick start for AI sessions:
ckb mcp --watch # Auto-reindexes every 30s when staleCheck staleness:
ckb status
# Shows: "5 commits behind HEAD" or "Up to date"For Go projects, CKB uses incremental indexing—only changed files are processed, making updates fast.
See the Index Management Guide for complete documentation.
Features
| Feature | Description |
|---|---|
| Compound Operations | explore, understand, prepareChange — single-call tools that reduce AI overhead by 60-70% |
| Code Navigation | Semantic search, call graphs, trace usage, find entrypoints |
| Impact Analysis | Blast radius, risk scoring, affected tests, breaking changes (compareAPI) |
| Architecture | Module overview, ADRs, dependency graphs, explain origin |
| Ownership | CODEOWNERS + git blame, reviewer suggestions, drift detection |
| Code Quality | Dead code detection (findDeadCode), coupling analysis, complexity |
| Security | Secret detection, credential scanning, allowlists |
| Documentation | Doc-symbol linking, staleness detection, coverage metrics |
| Multi-Repo | Federation, API contracts, remote index serving |
| Runtime | OpenTelemetry integration, observed usage, production dead code |
| Streaming | SSE streaming for findReferences, searchSymbols with real-time progress |
| Automation | Daemon mode, watch mode, webhooks, incremental indexing |
📖 Full Features Guide — Detailed documentation with examples
📋 Changelog — Version history
PR Review
CKB review runs 20 quality checks in 5 seconds — secrets, breaking changes, dead code, complexity, test gaps, bug patterns, and more. Zero tokens, zero API calls.
When your AI assistant (Claude Code, Cursor, Windsurf) reviews a PR, it calls CKB first and gets structured analysis in ~1k tokens. Then it only reads the files that matter — saving 50-80% of tokens on large PRs.
ckb review --base=main # Human-readable review
ckb review --base=main --ci # CI mode (exit codes)
ckb review --base=main --post=123 # Post as PR commentWorks in CI without any LLM:
- run: npx @tastehub/ckb review --base=main --ci --format=sarif > review.sarif| Without CKB | With CKB | |
|---|---|---|
| LLM tokens on 100-file PR | ~200k | ~50k |
| Files LLM reads | all | ~10 (CKB-flagged) |
| Secrets/breaking/dead-code checked | no | yes (all files) |
📖 How it helps AI review · Benchmarks · CI Integration · Quickstart
CLI
…