🧠 Codebase Curator

What is Codebase Curator?

Codebase Curator transforms how Claude understands your code by giving it a persistent companion that deeply analyzes your entire codebase. Using a two-Claude architecture, it spawns a dedicated “Curator Claude” that becomes an expert on your specific project, enabling fast, context-aware answers.

How to use Codebase Curator?

Install by cloning the repo, running bun install and bun link, then add the MCP server entry to your claude_code_config.json. In Claude Code, use the provided tools (e.g., set_project_path, get_codebase_overview, ask_curator, add_new_feature) to explore and ask questions. Alternatively, use the smartgrep CLI for semantic code search from any directory.

Key features of Codebase Curator

• Two-Claude architecture with persistent sessions and instant follow-ups
• Smart Grep semantic search with concept groups and advanced patterns
• Language support for 10 languages (TS/JS, Python, Go, Rust, Swift, Shell, JSON, YAML, TOML, .env)
• Live monitoring dashboard for real-time code evolution
• Incremental indexing with hash‑based change detection and debouncing
• Cross‑reference analysis showing definitions and usages

Use cases of Codebase Curator

• Get a deep overview of a new codebase in about two minutes
• Ask nuanced follow‑up questions that build on previous context
• Analyze the impact of uncommitted changes before committing
• Get guidance on where to add new features based on existing patterns
• Search semantic groups (e.g., all authentication patterns) with a single command

FAQ from Codebase Curator

How does Codebase Curator differ from asking Claude directly?

Instead of treating each question as isolated, Codebase Curator spawns a dedicated curator that becomes an expert on your codebase. It remembers context between questions, so follow‑ups are instant and build on prior understanding.

What are the runtime requirements?

You need an active Claude Code subscription and the Bun runtime installed. The integration uses the Model Context Protocol (MCP) to communicate between Claude and the curator.

Where is the codebase context stored?

The curator maintains persistent sessions. You can resume a session with the --resume flag. The README does not specify a remote storage; context is managed locally as part of the MCP server.

Is there a limit to the size of codebase it can handle?

The README mentions a streaming architecture that handles massive codebases efficiently and incremental indexing that only reprocesses changed files, but no hard size limit is stated.

What transports and authentication does it use?

Codebase Curator communicates via MCP (Model Context Protocol). No additional authentication is described; it operates within the Claude Code environment with your existing subscription.