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 that works with any LLM (including completely free local models via Ollama, or paid cloud models like Claude and gpt4.1) showcasing the power of this library in just a few lines of code.
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]git clone https://github.com/cased/kit.git
cd kit
uv venv .venv
source .venv/bin/activate
uv pip install -e .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}")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 and review a PR
kit review --init-config
kit review --dry-run https://github.com/owner/repo/pull/123
kit review https://github.com/owner/repo/pull/123The CLI supports all major repository operations with Unix-friendly output for scripting and automation. See the CLI Documentation for comprehensive usage examples.
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 at
a fraction of the cost with cloud models. At Cased we use kit extensively
with models like Sonnet 4 and gpt4.1, paying just for the price of tokens.
kit review --init-config
kit review https://github.com/owner/repo/pull/123Key Features:
- Whole repo context: Uses
kitso 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
kit also has first-class support for free local models via Ollama.
No API keys, no costs, no data leaving your machine.
📖 Complete PR Reviewer Documentation
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.
- High-level view with
-
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.
- Run regular expression searches across your codebase using
-
Prepare Code for LLMs & Analysis:
- Break down large files into manageable pieces for LLM context windows using
repo.chunk_file_by_lines()orrepo.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().
- Break down large files into manageable pieces for LLM context windows using
-
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()). - Works with any LLM: free local models (Ollama), or cloud models (OpenAI, Anthropic, Google).
- Build a searchable semantic index of these AI-generated docstrings with
DocstringIndexerand query it withSummarySearcherto find code based on intent and meaning.
- Use LLMs to create natural language summaries for files, functions, or classes using the
-
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()andanalyzer.generate_llm_context().
- Map import relationships between modules using
-
Repository Versioning & Historical Analysis:
- Analyze repositories at specific commits, tags, or branches using the
refparameter. - 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.
- Analyze repositories at specific commits, tags, or branches using the
-
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 reviewusing free local models (Ollama) or cloud models (Claude, GPT-4). - 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 (free for Ollama).
- Automated PR review with
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:
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.
Explore the Full Documentation for detailed usage, advanced features, and practical examples. Full REST documentation is also available.
📝 Changelog - Track all changes and improvements across kit releases
MIT License
- 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.
- Discord: Join the Discord to talk kit and Cased
To contribute, fork the repository, make your changes, and submit a pull request.
{ "mcpServers": { "kit-mcp": { "command": "python", "args": ["-m", "kit.mcp"] } } }