TL;DR: Today we're launching Context Engine MCP to bring Augment's industry-leading semantic search to every MCP-compatible agent. In our benchmarks, adding Context Engine improved agent performance by 70%+ across Claude Code, Cursor, and Codex. Whether you use these or any other MCP-compatible agent, you can now give it deep codebase context that makes it write better code, faster, and for fewer tokens. Every Augment Code user will get 1,000 requests for free in February.
Set it up in 2 minutes →
Try it yourself
Context Engine MCP now works with every major AI coding agent.
The Context Engine MCP is available today for any MCP-compatible agent.
Point it at your repository, and your agents immediately gain semantic understanding of your codebase architecture, patterns, and conventions.
Context Engine MCP supports two modes:
- Local mode (recommended for active development): Run
AuggieCLI locally as an MCP server. Your agent retrieves context from your working directory in real-time. - Remote mode (recommended for cross-repo context): Connect to Augment's hosted MCP at https://api.augmentcode.com/mcp for context across all your repositories.
Setup takes about 2 minutes for either mode.
Context is the problem
Every engineering team using AI coding agents faces the same problems: agents hallucinate, waste tokens on irrelevant context, and burn budget making the same mistakes repeatedly. The root cause? Models lack deep understanding of your codebase architecture, dependencies, and patterns.
We built our Context Engine to solve this. By exposing it through Model Context Protocol (MCP), any agent, including Cursor, Claude Code, and Codex, can tap into semantic search that makes models meaningfully better.
Here's what happened when we ran controlled evaluations.
The benchmark
We evaluated agents on a real-world task: take a natural language prompt and ship a complete PR. The dataset: 300 Elasticsearch PRs, each with 3 different prompts, for 900 attempts total.
We measured five dimensions that matter for production code: correctness, completeness, best practices, code reuse, and unsolicited documentation.
Results
Augment's Context Engine MCP delivers 30–80% quality improvements across leading AI coding agents, with faster completion times and lower token costs.
The Context Engine MCP delivered consistent quality gains regardless of agent or model:
- Cursor + Claude Opus 4.5: 71% improvement (completeness +60%, correctness +5x)
- Claude Code + Opus 4.5: 80% improvement
- Cursor + Composer-1: 30% improvement, bringing a struggling model into viable territory
Faster and cheaper, not just better
You'd expect higher quality to cost more tokens. The opposite happened. The Context Engine helps models find the right answer faster—fewer exploratory turns, less backtracking, more targeted code generation. MCP-enabled runs consistently required fewer tool calls and conversation turns. Code that's more correct from the start means less time debugging what the agent got wrong.
Context architecture is as important as model choice.
Most "AI code quality" discussions focus on model selection: should I use Opus or Sonnet? GPT-5 or Gemini? Our data shows context architecture matters as much or more than model choice.
A weaker model with great context (Sonnet + MCP) can outperform a stronger model with poor context (Opus without MCP). And when you give the best models great context, they deliver step-function improvements in production-ready code quality.
The Context Engine MCP works because it provides:
- Semantic search: Not just text search, but comprehension of relationships, dependencies, and architectural patterns
- Precise context selection: Surface exactly what's relevant to the task, nothing more
- Universal integration: Works with any MCP-compatible agent through an open protocol
Better code, lower cost, faster iteration. That's what happens when models actually understand your codebase. Try it for yourself today.
Written by
Sylvain Giuliani
Sylvain Giuliani is the head of growth at Augment Code, leveraging more than a decade of experience scaling developer-focused SaaS companies from $1 M to $100 M+ ARR. Before Augment, he built and led the go-to-market and operations engines at Census and served as CRO at Pusher, translating deep data insights into outsized revenue gains.