LeanKG

Lightweight Knowledge Graph for AI-Assisted Development

LeanKG is a local-first knowledge graph that gives AI coding tools accurate codebase context. It indexes your code, builds dependency graphs, generates documentation, and exposes an MCP server so tools like Cursor, OpenCode, and Claude Code can query the knowledge graph directly. No cloud services, no external databases -- everything runs on your machine with minimal resources.

Token Savings Example (Benchmarked)

Real benchmark results from the Go API Service example:

Scenario	Without LeanKG	With LeanKG	Savings
Impact Analysis	835 tokens	13 tokens	98.4%
Full Feature Testing	9,601 tokens	42 tokens	99.6%

# Run the benchmark yourself
cd examples/go-api-service
python3 benchmark.py

Before LeanKG: AI must scan entire codebase to understand dependencies (~9,600 tokens)

After LeanKG: LeanKG provides targeted subgraph with relationships pre-computed (~42 tokens)

Why LeanKG?

AI coding tools waste tokens scanning entire codebases. LeanKG provides targeted context instead:

Scenario	Without LeanKG	With LeanKG
File review	Full content of changed files + diff	Blast radius + structural summary
Impact analysis	Manually trace dependencies	`get_impact_radius` returns affected files
Token count	9,600+ tokens for full scan	13-42 tokens with graph

Installation

Quick Install via npm (Recommended -- No Rust Required)

npm install -g leankg
leankg --version

The npm package downloads pre-built binaries for your platform. Supported: macOS (x64, ARM64), Linux (x64, ARM64).

Install via Cargo

cargo install leankg
leankg --version

One-Line Install (Shell Script)

Install the LeanKG binary and configure MCP for your AI coding tool:

curl -fsSL https://raw.githubusercontent.com/FreePeak/LeanKG/main/scripts/install.sh | bash -s -- <target>

Supported targets:

Target	AI Tool
`opencode`	OpenCode AI
`cursor`	Cursor AI
`claude`	Claude Code/Desktop
`gemini`	Gemini CLI
`antigravity`	Anti Gravity

Examples:

# Install for OpenCode
curl -fsSL https://raw.githubusercontent.com/FreePeak/LeanKG/main/scripts/install.sh | bash -s -- opencode

# Install for Cursor
curl -fsSL https://raw.githubusercontent.com/FreePeak/LeanKG/main/scripts/install.sh | bash -s -- cursor

# Install for Claude Code
curl -fsSL https://raw.githubusercontent.com/FreePeak/LeanKG/main/scripts/install.sh | bash -s -- claude

Build from Source

git clone https://github.com/your-org/LeanKG.git
cd LeanKG
cargo build --release

Quick Start

# 1. Initialize LeanKG in your project
leankg init

# 2. Index your codebase
leankg index ./src

# 3. Start the MCP server (for AI tools)
leankg serve

# 4. Compute impact radius for a file
leankg impact src/main.rs --depth 3

# 5. Check index status
leankg status

How It Works

sequenceDiagram
    participant Dev as Developer
    participant CLI as LeanKG CLI
    participant Indexer as Code Indexer
    participant DB as CozoDB
    participant MCP as MCP Server
    participant AI as AI Tool (Claude/Cursor)

    Dev->>CLI: leankg init
    CLI->>DB: Initialize graph database

    Dev->>CLI: leankg index ./src
    CLI->>Indexer: Parse source files
    Indexer->>Indexer: Extract functions, imports, calls
    Indexer->>DB: Store code elements & relationships

    Dev->>CLI: leankg serve
    CLI->>MCP: Start MCP server

    AI->>MCP: "What's the impact of changing auth.rs?"
    MCP->>DB: Query impact radius (N hops)
    DB-->>MCP: Affected files list
    MCP-->>AI: Targeted context (13 tokens vs 835)

    Dev->>CLI: leankg watch
    CLI->>Index: Watch for file changes
    Index->>DB: Incremental update

Index -- LeanKG parses your codebase and builds a graph of code elements (functions, classes, modules) and their relationships (imports, calls, tests).
Query -- AI tools query the graph via MCP instead of scanning files.
Optimize -- Get targeted context with ~99% token reduction.

MCP Server Setup

LeanKG exposes a Model Context Protocol (MCP) server that AI tools can connect to.

Option 1: Automated Setup (Recommended)

leankg install

Detects your AI tool (Claude Code, OpenCode, Cursor, etc.) and installs the appropriate MCP configuration.

Option 2: Manual Setup

Claude Code / Claude Desktop

Add to ~/.config/claude/settings.json:

{
  "mcpServers": {
    "leankg": {
      "command": "leankg",
      "args": ["mcp-stdio", "--watch"]
    }
  }
}

Cursor

Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "leankg": {
      "command": "leankg",
      "args": ["mcp-stdio", "--watch"]
    }
  }
}

OpenCode

Add to ~/.opencode/mcp.json:

{
  "mcpServers": {
    "leankg": {
      "command": "leankg",
      "args": ["mcp-stdio", "--watch"]
    }
  }
}

Starting the MCP Server

# Stdio mode with auto-indexing (for local AI tools)
leankg mcp-stdio --watch

# Stdio mode without auto-indexing
leankg mcp-stdio

Highlights

Code Indexing -- Parse and index Go, TypeScript, Python, and Rust codebases with tree-sitter.
Dependency Graph -- Build call graphs with IMPORTS, CALLS, and TESTED_BY edges.
Impact Radius -- Compute blast radius for any file to see downstream impact.
Auto Documentation -- Generate markdown docs from code structure automatically.
MCP Server -- Expose the graph via MCP protocol for AI tool integration.
File Watching -- Watch for changes and incrementally update the index.
CLI -- Single binary with init, index, serve, impact, and status commands.
Business Logic Mapping -- Annotate code elements with business logic descriptions and link to features.
Traceability -- Show feature-to-code and requirement-to-code traceability chains.
Documentation Mapping -- Index docs/ directory, map doc references to code elements.

Auto-Indexing

LeanKG watches your codebase and automatically keeps the knowledge graph up-to-date.

# Start file watcher -- indexes changes automatically in background
leankg watch

# Incremental indexing -- only re-index changed files (git-based)
leankg index --incremental

# Filter by language
leankg index --lang go,ts,py,rs

# Exclude patterns
leankg index --exclude vendor,node_modules,dist

graph LR
    subgraph "File Watcher"
        FS[File System Events]
        Git[Git Status]
        Parse[Parser]
        DB[(CozoDB)]
    end

    FS -->|change detected| Git
    Git -->|only changed files| Parse
    Parse -->|update relationships| DB

Watch Mode -- leankg watch monitors your source directory for file changes.
Git-Based Delta -- Uses git diff to detect only modified files.
Incremental Update -- Re-parses only changed files and updates affected relationships.
Background Sync -- Runs in background while you code.

Architecture

graph TB
    subgraph "AI Tools"
        Claude[Claude Code]
        Open[OpenCode]
        Cursor[Cursor]
        Antigravity[Google Antigravity]
    end

    subgraph "LeanKG"
        CLI[CLI Interface]
        MCP[MCP Server]
        Watcher[File Watcher]

        subgraph "Core"
            Indexer[tree-sitter Parser]
            Graph[Graph Engine]
            Cache[Query Cache]
        end

        subgraph "Storage"
            CozoDB[(CozoDB)]
        end

        Web[Web UI]
    end

    Claude --> MCP
    Open --> MCP
    Cursor --> MCP
    Antigravity --> MCP
    CLI --> Indexer
    CLI --> Graph
    Watcher --> Indexer
    Indexer --> CozoDB
    Graph --> CozoDB
    Graph --> Cache
    Web --> Graph

CLI Commands

Command	Description
`leankg init`	Initialize LeanKG in the current directory
`leankg index [path]`	Index source files at the given path
`leankg index --incremental`	Only index changed files (git-based)
`leankg index --lang go,ts,py,rs`	Filter by language
`leankg index --exclude vendor,node_modules`	Exclude patterns
`leankg serve`	Start the MCP server (WebSocket)
`leankg serve --mcp-port 3000`	Custom MCP server port
`leankg mcp-stdio`	Start MCP server with stdio transport
`leankg impact <file> --depth N`	Compute blast radius for a file
`leankg status`	Show index statistics and status
`leankg generate`	Generate documentation from the graph
`leankg install`	Auto-install MCP config for AI tools
`leankg watch`	Start file watcher for auto-indexing
`leankg quality --min-lines N`	Find oversized functions by line count
`leankg query <text> --kind name`	Query the knowledge graph
`leankg annotate <element> -d <desc>`	Add business logic annotation
`leankg link <element> <id>`	Link element to feature
`leankg search-annotations <query>`	Search business logic annotations
`leankg show-annotations <element>`	Show annotations for a specific element
`leankg trace --feature <id>`	Show feature-to-code traceability
`leankg find-by-domain <domain>`	Find code by business domain
`leankg export`	Export graph data as JSON
`leankg docs --tree`	Show documentation directory structure
`leankg docs --for <file>`	Show docs referencing a code file
`leankg docs --link <doc> <element>`	Link documentation to code element
`leankg trace <element>`	Show traceability chain for element
`leankg trace --requirement <id>`	Trace code for a requirement

MCP Tools

Tool	Description
`query_file`	Find file by name or pattern
`get_dependencies`	Get file dependencies (direct imports)
`get_dependents`	Get files depending on target
`get_impact_radius`	Get all files affected by change within N hops
`get_review_context`	Generate focused subgraph + structured review prompt
`get_context`	Get AI context for file (minimal, token-optimized)
`find_function`	Locate function definition
`get_call_graph`	Get function call chain (full depth)
`search_code`	Search code elements by name/type
`generate_doc`	Generate documentation for file
`find_large_functions`	Find oversized functions by line count
`get_tested_by`	Get test coverage for a function/file
`get_doc_for_file`	Get documentation files referencing a code element
`get_files_for_doc`	Get code elements referenced in a documentation file
`get_doc_structure`	Get documentation directory structure
`get_traceability`	Get full traceability chain for a code element
`search_by_requirement`	Find code elements related to a requirement
`get_doc_tree`	Get documentation tree structure
`get_code_tree`	Get codebase structure
`find_related_docs`	Find documentation related to a code change

Supported AI Tools

Tool	Integration	Status
Claude Code	MCP	Supported
OpenCode	MCP	Supported
Cursor	MCP	Supported
Google Antigravity	MCP	Supported
Windsurf	MCP	Supported
Codex	MCP	Supported

Roadmap

Phase 2 -- Pipeline Integration

Feature	Status	Description
Pipeline Parsing	Planned	Parse CI/CD config files (GitHub Actions, GitLab CI, Jenkins, Azure)
Pipeline Graph	Planned	Build pipeline, stage, step nodes
Trigger Links	Planned	Link source file changes to triggered pipelines
Pipeline Impact	Planned	Include pipelines in blast radius analysis
Deployment Targets	Planned	Track which stages deploy to which environments

Supported CI/CD Platforms (Coming Soon):

GitHub Actions (.github/workflows/*.yml)
GitLab CI (.gitlab-ci.yml)
Jenkins (Jenkinsfile)
Azure Pipelines (azure-pipelines.yml)

Future Features

Feature	Description
Semantic Search	AI-powered code search using embeddings
Security Analysis	Detect vulnerable dependencies and patterns
Cost Estimation	Cloud resource cost tracking via pipeline data
Multi-Project	Index and query across multiple repositories

Requirements

For npm installation (recommended):

Node.js 18+
npm 8+

For building from source:

Rust 1.70+
macOS or Linux

Tech Stack

Component	Technology
Language	Rust
Database	CozoDB (embedded relational-graph, Datalog queries)
Parsing	tree-sitter
CLI	Clap
Web Server	Axum
Installer	Node.js (npm package for binary distribution)

Project Structure

src/
  cli/         - CLI commands (Clap)
  config/      - Project configuration
  db/          - CozoDB persistence layer
  doc/         - Documentation generator
  graph/       - Graph query engine
  indexer/     - Code parser (tree-sitter)
  doc_indexer/ - Documentation indexer
  mcp/         - MCP protocol handler
  watcher/     - File change watcher
  web/         - Web server (Axum)

docs/
  planning/    - Planning documents
  requirement/ - Requirements documents (PRD)
  analysis/    - Analysis documents
  design/      - Design documents (HLD)
  business/    - Business logic documents

License

MIT

leankg 0.2.0