Ckb Code Knowledge Backend

What is CKB?

CKB (Code Knowledge Backend) transforms your codebase into a queryable knowledge base, enabling developers, AI assistants, and CI/CD pipelines to understand, analyze, protect, and improve code. It provides semantic search, impact analysis, ownership lookup, dead code detection, and secret scanning through a CLI, HTTP API, or MCP (Model Context Protocol) server.

How to use CKB?

Install via npm (npm install -g @tastehub/ckb) or Homebrew (brew install ckb). Initialize in a project with ckb init, optionally generate a SCIP index with ckb index, and connect to AI assistants like Claude Code using ckb setup. Use CLI commands (ckb search, ckb hotspots), start the HTTP API with ckb serve, or run the MCP server with ckb mcp for integration with AI tools.

Key features of CKB

• Semantic search, call graphs, usage tracing, and architecture maps
• Impact analysis, risk scoring, hotspot detection, and coupling analysis
• Affected test detection, breaking change warnings, and PR risk assessment
• Secret detection, credential scanning, and security-sensitive code identification
• Ownership lookup, reviewer suggestions, and architectural decision records
• Dead code detection, tech debt tracking, and documentation coverage

Use cases of CKB

• Assess what breaks when changing a function, with precise blast radius and risk scores
• Suggest reviewers for a PR using data-driven ownership and drift detection
• Detect dead code with confidence scores and find legacy code that can be removed
• Automatically identify which tests to run after a change, reducing CI time from 30 to 2 minutes
• Scan codebase for exposed secrets using 26 patterns and automated scanning

FAQ from CKB

What languages does CKB support?

CKB classifies languages into quality tiers: Tier 1 (Go – full support), Tier 2 (TypeScript, JavaScript, Python – full support with minor edge cases), Tier 3 (Rust, Java, Kotlin, C++, Ruby, Dart – basic support), and Tier 4 (C#, PHP – experimental). Incremental indexing is Go-only.

How does indexing work?

CKB uses SCIP indexes to understand code structure. Run ckb index to generate an index. Without an index, CKB works via tree-sitter parsing (basic mode). With an index, you get cross-file references, precise impact analysis, and call graph navigation. Indexes become stale after commits; refresh manually, with watch mode (ckb index --watch), or via CI webhooks.

How do I connect CKB to an AI assistant?

Run ckb setup to automatically create a .mcp.json configuration for Claude Code. Alternatively, manually add the MCP server: claude mcp add --transport stdio ckb -- npx @tastehub/ckb mcp. Once connected, you can ask questions like "What calls the HandleRequest function?" or "Who owns the internal/api module?"

What is the token efficiency of CKB's MCP server?

CKB uses smart presets that load only the tools you need, achieving up to 83% token reduction. At startup, it reports active tools (e.g., 14 of 76) and estimated context (~1k tokens). Its compound operations (explore, understand, prepareChange) reduce AI tool calls by 60-70%.