Ark

•
25
•

Extracts code declarations, resolves references, and provides structural graph intelligence to AI coding agents via the Model Context Protocol.

Ark is a high-performance code intelligence engine engineered specifically for AI coding agents. Delivered as a single static binary, it parses multi-language repositories using Tree-sitter without running project code, extracting declarations and cross-file references into a structured dependency graph. Instead of forcing AI assistants to consume excessive tokens reading entire files or guessing ambiguous names via blind text search, Ark exposes 21 MCP tools that trace callers, callees, and change impact with explicit confidence ratings. By providing token-budgeted, evidence-backed context, it enables agents like Claude, Cursor, and Copilot to navigate complex codebases safely and efficiently.

주요 기능

01Zero-execution architecture that reads code statically without outbound network calls or external builds
02Context budgeting that delivers focused declarations and references instead of whole files
03Structural code intelligence supporting Go, TypeScript, Java, Rust, Python, PHP, Zig, and configuration formats
0425 GitHub stars
05Evidence-based resolution engine rating relationship certainty as exact, strong, or candidate
06Change impact analysis reporting direct, transitive, and test dependents before code modifications

사용 사례

01Tracing call graphs and symbol dependencies across packages without relying on broad text searches
02Providing token-optimized symbol context and directory architecture maps to LLM coding assistants
03Analyzing the blast radius and potential breaking changes across a repository prior to refactoring