Augment opens Context Engine via MCP

This is less a flashy model launch than a power move on the agent infrastructure layer: Augment wants to be the retrieval spine underneath the rest of the coding-agent market. If its evals generalize, MCP makes context quality portable instead of locking it inside one IDE.

• –The launch adds both a local MCP server via Auggie CLI for live working-directory indexing and a hosted remote MCP endpoint for cross-repo context
• –Augment’s docs already ship quickstarts for Cursor, Claude Code, Codex, GitHub Copilot, Gemini CLI, Zed, Kiro, Roo Code, and more, which broadens reach beyond its own app
• –The company claims 30–80% quality gains on 900 Elasticsearch task runs, plus fewer tool calls and lower token costs, which is a strong developer-value pitch if replicated independently
• –The real message is strategic: model quality is becoming commoditized, while context architecture is emerging as the durable advantage in AI coding