styler-ai/ProjectAtlas

What it solves

ProjectAtlas addresses the high token cost and inefficiency associated with coding agents (like Claude Code or Codex) when navigating large codebases. Instead of agents reading entire directories or irrelevant files to find the right code, ProjectAtlas provides a local, intelligent map that guides them to the exact source slices they need, significantly reducing token consumption.

How it works

It is a Rust-native CLI and Model Context Protocol (MCP) server that creates a persistent SQLite database within the project directory (.projectatlas/). It scans the repository—respecting .gitignore—to index symbols, graph relationships, folder/file purposes, and deterministic summaries. Agents use MCP tools to query this map (e.g., via atlas_session_brief) to narrow down the relevant code before reading the actual source files.

Who it’s for

Developers using AI coding agents (such as Codex, Claude Code, or OpenCode) who work in large repositories and want to lower their LLM token costs and improve agent accuracy.

Highlights

• High Performance: Built in Rust for fast indexing and low-latency lookups.
• Token Efficiency: Aims to reduce token costs by providing compact summaries and targeted source slices rather than broad file reads.
• MCP Integration: Native support for the Model Context Protocol, allowing seamless integration with compatible AI agents.
• Git Worktree Support: Ability to track and route across multiple Git worktrees using short aliases.
• Local First: No hosted index or credentials required; all state is stored locally in SQLite.