CodeGraph: Pre-indexed code knowledge graph for coding agents, runs locally and auto-syncs

CodeGraph pre-indexes the whole repository into a knowledge graph: every symbol, call edge and dependency, including dynamic dispatches that grep can't follow. The parsing core is written in Rust and covers 20 languages, including TypeScript, JavaScript, Java, Python, Go, C, C++, Rust, C#, Ruby, PHP, Swift, Kotlin, Scala, Dart, R, Lua and Luau. Install with one curl or PowerShell command, no Node.js required; codegraph install writes the MCP server config into Claude Code, Cursor, Codex CLI, opencode, Hermes Agent, Gemini CLI, Antigravity IDE, Kiro and GitHub Copilot; codegraph init builds the .codegraph/ directory and indexes once, then watches by default so the graph updates automatically on file changes. The index stays local. Official comparisons across 7 open-source repos (VS Code, Excalidraw, Django, Tokio, OkHttp, Gin, Alamofire) show on average 88% fewer tool calls, 53% faster, 62% fewer tokens, 44% cheaper, and file reads dropping to zero; on VS Code, 2m10s down to 58s with 77% fewer tokens. Tests ran Claude Opus 4.8 headless, with the codegraph CLI blocked on both arms to avoid contamination. That same mechanism has a downside: CodeGraph returns dense raw snippets that linger in the context window. Over multi-turn sessions it leaves roughly 80% more residual retrieval context than file-by-file agents—67k vs 18k tokens on VS Code. Developers on smaller context windows should budget long sessions carefully, while teams tracing call chains across files every query see direct savings in tool calls and cost.