Berkelium CLI

Berkelium CLI is a Code Graph Management tool and Model Context Protocol (MCP) server. It uses tree-sitter to parse your codebase into a structured graph stored in SQLite, enabling high-fidelity impact analysis and surgical context retrieval for AI assistants.

Quick Start

Installation

# Install
pip install berkelium

# Run the TUI
berkelium-cli

Development Setup

git clone https://github.com/BerkeliumLabs/berkelium-cli
cd berkelium-cli
uv sync

TUI Features

The Berkelium TUI (berkelium-cli) provides a terminal interface for:

• Build/Update Graph: Index supported languages (Python, JS/TS, Go, Java, Rust, C/C++).
• Incremental Sync: Uses git-diff to update the graph in milliseconds.
• Exploration: Visualize symbols and relationships directly in your terminal.

MCP Server (for AI Assistants)

Connect Berkelium to Claude, Cursor, or any MCP-compatible client to give your AI "graph-vision" over your code.

Configuration (Claude)

Add this to your settings.json:

{
  "mcpServers": {
    "berkelium": {
      "command": "berkelium-mcp",
    }
  }
}

Available Tools

• query_search_codebase: Executes read-only Cypher queries directly against the code graph for advanced structural search.

Available Prompts

• review_my_pr: A guided workflow that syncs the graph, identifies changes via git, and suggests targeted tests based on impact analysis.

Usage Examples (AI-Assisted Development)

Berkelium is most powerful when used through an AI assistant (like Claude or Cursor). Here are some real-world scenarios where it excels:

1. Architectural Exploration

Prompt: "I want to add a new database provider. Show me all classes that inherit from BaseProvider to see the existing pattern."

• AI Action: Uses query_search_codebase with a Cypher query like:

  MATCH (child)-[:INHERITS]->(parent) WHERE parent.name = 'BaseProvider' RETURN child.name AS name, child.file_rel_path AS file
  

2. Tracing Complex Call Chains

Prompt: "Trace the execution flow starting from IncrementalSync.sync(). What internal methods does it reach?"

• AI Action: Uses query_search_codebase to traverse the [:CALLS] edges.

3. Identifying Dead Code

Prompt: "Are there any functions in the utils/ directory that are never called by any other part of the project?"

• AI Action: Uses query_search_codebase to find nodes with zero incoming [:CALLS] edges.

Architecture

• Extractor (extractor.py): Language-agnostic extraction using tree-sitter.
• Store (store.py): Persistence layer using GraphQLite on SQLite.
• Sync (sync.py): Git-based incremental synchronization logic.
• Retriever (retriever.py): Graph traversal algorithms for impact and context.

License

MIT License - see LICENSE.