AI coding startup Augment Code has unveiled a “model context protocol” for AI agents and platforms, SiliconANGLE reported on Thursday.
The company said the update improves accuracy by letting AI understand a codebase through semantic analysis rather than simple keyword search. In its own benchmark, it recorded performance gains of more than 70 percent when combined with Claude Code, Cursor and Codex, among others.
Existing AI coding agents have had problems generating incorrect results or consuming unnecessary tokens because they do not fully grasp code structure and dependencies. Augment Code said its context engine helps AI understand a codebase’s architecture and patterns, improving code quality by 30 to 80 percent. The company said it confirmed a 71 percent improvement with the Cursor plus Claude Opus 4.5 combination and an 80 percent improvement with Claude Code plus Opus 4.5.
The context engine supports AI models in generating high-quality code with fewer resources. For example, the company said Sonnet, a smaller model, can deliver better performance than a large model such as Opus when combined with the context engine.
Chi-gyu Hwang
delight@d-today.co.kr