cased-kit 0.5.0

A modular toolkit for LLM-powered codebase understanding.

kit 🛠️ Code Intelligence Toolkit

kit is a production-ready toolkit for codebase mapping, symbol extraction, code search, and building LLM-powered developer tools, agents, and workflows.

Use kit to build things like code reviewers, code generators, even IDEs, all enriched with the right code context.

Work with kit directly from Python, or with MCP + function calling, REST, or CLI!

kit also ships with damn fine PR reviewer for free (just pay for LLM usage), showcasing the power of this library in just a few lines of code.

Quick Installation

Install from PyPI

pip install cased-kit

# With semantic search features (includes PyTorch, sentence-transformers)
pip install cased-kit[ml]

# Everything (including MCP server and all features)
pip install cased-kit[all]

Install from Source

git clone https://github.com/cased/kit.git
cd kit
uv venv .venv
source .venv/bin/activate
uv pip install -e .

Basic Usage

Python API

from kit import Repository

# Load a local repository
repo = Repository("/path/to/your/local/codebase")

# Load a remote public GitHub repo
# repo = Repository("https://github.com/owner/repo")

# Load a repository at a specific commit, tag, or branch
# repo = Repository("https://github.com/owner/repo", ref="v1.2.3")

# Explore the repo
print(repo.get_file_tree())
# Output: [{"path": "src/main.py", "is_dir": False, ...}, ...]

print(repo.extract_symbols('src/main.py'))
# Output: [{"name": "main", "type": "function", "file": "src/main.py", ...}, ...]

# Access git metadata
print(f"Current SHA: {repo.current_sha}")
print(f"Branch: {repo.current_branch}")

Command Line Interface

kit also provides a comprehensive CLI for repository analysis and code exploration:

# Get repository file structure
kit file-tree /path/to/repo

# Extract symbols (functions, classes, etc.)
kit symbols /path/to/repo --format table

# Search for code patterns
kit search /path/to/repo "def main" --pattern "*.py"

# Find symbol usages
kit usages /path/to/repo "MyClass"

# Export data for external tools
kit export /path/to/repo symbols symbols.json

# Initialize configuration
kit review --init-config

# Review a PR
kit review https://github.com/owner/repo/pull/123
kit review --dry-run https://github.com/owner/repo/pull/123

The CLI supports all major repository operations with Unix-friendly output for scripting and automation. See the CLI Documentation for comprehensive usage examples.

AI-Powered PR Reviews

As both of a demonstration of this library, and as a standalone product, kit includes a MIT-licensed, CLI-based pull request reviewer that ranks with the better closed-source paid options—but for a fraction of the cost. Have thousands of PRs reviewed at cost, say $20 per month for all of them. A typical high-quality kit review with a SOTA model costs about 10 cents.

# Configure your LLM provider (Claude or GPT-4)
kit review --init-config

# Review a PR
kit review https://github.com/owner/repo/pull/123

Key Features:

• Whole repo context: Uses kit so has all the features of this library
• Production-ready: Rivals paid services, but MIT-licensed; just pay for tokens
• Cost transparency: Real-time token usage and pricing
• Fast: No queuing, shared services: just your code and the LLM
• Works from wherever: Trigger reviews with the CLI, or run it via CI

📖 Complete PR Reviewer Documentation

Key Features & Capabilities

kit helps your apps and agents understand and interact with codebases, with components to build your own AI-powered developer tools.

• Explore Code Structure:

  • High-level view with repo.get_file_tree() to list all files and directories.
  • Dive down with repo.extract_symbols() to identify functions, classes, and other code constructs, either across the entire repository or within a single file.

• Pinpoint Information:

  • Run regular expression searches across your codebase using repo.search_text().
  • Track specific symbols (like a function or class) with repo.find_symbol_usages().
  • Perform semantic code search using vector embeddings to find code based on meaning rather than just keywords.

• Prepare Code for LLMs & Analysis:

  • Break down large files into manageable pieces for LLM context windows using repo.chunk_file_by_lines() or repo.chunk_file_by_symbols().
  • Get the full definition of a function or class off a line number within it using repo.extract_context_around_line().

• Generate Code Summaries:

  • Use LLMs to create natural language summaries for files, functions, or classes using the Summarizer (e.g., summarizer.summarize_file(), summarizer.summarize_function()).
  • Build a searchable semantic index of these AI-generated docstrings with DocstringIndexer and query it with SummarySearcher to find code based on intent and meaning.

• Analyze Code Dependencies:

  • Map import relationships between modules using repo.get_dependency_analyzer() to understand your codebase structure.
  • Generate dependency reports and LLM-friendly context with analyzer.generate_dependency_report() and analyzer.generate_llm_context().

• Repository Versioning & Historical Analysis:

  • Analyze repositories at specific commits, tags, or branches using the ref parameter.
  • Compare code evolution over time, work with diffs, ensure reproducible analysis results
  • Access git metadata including current SHA, branch, and remote URL with repo.current_sha, repo.current_branch, etc.

• Multiple Access Methods:

  • Python API: Direct integration for building applications and scripts.
  • Command Line Interface: 11+ commands for shell scripting, CI/CD, and automation workflows.
  • REST API: HTTP endpoints for web applications and microservices.
  • MCP Server: Model Context Protocol integration for AI agents and development tools.

• AI-Powered Code Review:

  • Automated PR review with kit review using Claude or GPT-4 for intelligent code analysis.
  • Repository cloning and comprehensive file analysis for deep code understanding.
  • Configurable review depth (quick, standard, thorough) and customizable analysis settings.
  • Seamless GitHub integration with automatic comment posting and PR workflow integration.
  • Cost transparency with real-time LLM token usage tracking and pricing information.

MCP Server

The kit tool includes an MCP (Model Context Protocol) server that allows AI agents and other development tools to interact with a codebase programmatically.

MCP support is currently in alpha. Add a stanza like this to your MCP tool:

{
  "mcpServers": {
    "kit-mcp": {
      "command": "python",
      "args": ["-m", "kit.mcp"]
    }
  }
}

The python executable invoked must be the one where cased-kit is installed. If you see ModuleNotFoundError: No module named 'kit', ensure the Python interpreter your MCP client is using is the correct one.

Documentation

Explore the Full Documentation for detailed usage, advanced features, and practical examples. Full REST documentation is also available.

License

MIT License

Contributing

• Local Development: Check out our Running Tests guide to get started with local development.
• Project Direction: See our Roadmap for future plans and focus areas.

To contribute, fork the repository, make your changes, and submit a pull request.