The Agent Formula: PE + Tools
At Trae, we've developed a powerful framework understanding that Agent = PE (Prompt Engineering) + Tools. This isn't just theoretical—it's a practical architecture that drives real results:
-
Prompt Engineering provides the Agent with clear context, goals, and constraints to understand specific work domains
-
Tools supply the necessary capabilities and resources to execute specialized tasks effectively
Build Your Own Customized Agents
Beyond Trae's built-in Agents, we're empowering developers to create custom Agents tailored to their unique workflows. The process is straightforward yet powerful:
-
Define Rules that shape Agent behavior—from coding style preferences to complexity limits and architectural guidelines
-
Select Tools that determine which resources your Agent can access—choose between different models, APIs, and capabilities
We're introducing Rules with @Agent because we recognize that Rules aren't just for prompt engineering—they're fundamental building blocks for Agent behavior itself. This integration creates a flexible system where developers can craft specialized AI collaborators that truly understand their unique workflows and requirements.
Experience the power of a development team that adapts to your needs through simple, intuitive prompts—the future of Human-AI collaboration is here.
MCP Integration: The Neural Network Connecting Your Agents
This is just the beginning of @Agent's potential. We're integrating MCP (Model Context Protocol) into Trae, establishing a universal communication framework that enables seamless interactions between Agents and third-party extensions. MCP intelligently coordinates the optimal combination of Agents based on your specific requirements, orchestrating them into a unified problem-solving team.
An Agent-First Architecture
@Agent is our core capability. We let @Agent #Context deliver "user query".
Trae takes a different approach on MCP and Agents. We've designed an agent-first architecture where MCP serves as the underlying coordination layer rather than a user-facing interface. This architectural choice makes complex AI collaboration more intuitive and results-oriented.
By packaging MCP, local tools, and contextual information within each Agent, we create specialized AI collaborators that autonomously handle the right combination of Rules and Tools for each circumstance. This modular design allows each Agent to focus on their specific domain expertise without unnecessary complexity.
The result is a more coherent development experience where you interact with purpose-built Agents rather than managing the underlying protocols directly. This approach delivers the power of multi-agent coordination without the cognitive overhead, keeping your workflow focused on creation rather than configuration.
Building an Open Agent Ecosystem
We're creating more than a product—we're establishing a thriving ecosystem where infinite intelligence emerges through collaboration. The true power of Trae extends beyond our built-in capabilities to the collective innovation of our developer community.
From Individual Creation to Collective Intelligence
We're inviting developers and builders to actively contribute by creating customized Agents that expand Trae's capabilities exponentially. Our vision is an AI programming ecosystem where users freely create, share, and utilize diverse Agents—each contribution adding to a network of capabilities accessible to everyone.
Together, we're building toward a more efficient, open, and intelligent programming future—one where the boundaries between what you can imagine and what you can create continue to dissolve.