Agents

What Are Agents?

Agents are single-purpose virtual MCPs optimized for specific tasks. While Projects provide general-purpose work environments, agents focus narrowly on accomplishing one thing well.

Agents provide:

• Focused capabilities - Curated tools for specific tasks
• Task-specific context - Resources and guidelines tailored to the agent’s purpose
• Single-purpose design - Clear objective with no unnecessary complexity

Agents are organization-wide configurations available in the sidebar under the Agents section. Each agent can be spawned in any scope (tasks, subtasks, projects) with fresh execution context, providing a focused workspace for its specific task.

The key distinction: projects organize capabilities for broad work contexts, while agents optimize for single-purpose execution.

How Agents Work

Agents follow the same pattern as standard agent skills ( SKILL.md files), making them familiar and easy to understand:

Like SKILL.md skills:

• Description - Brief explanation of what the agent does and when to use it (like the description: field in SKILL.md )
• Instructions - Detailed guidelines for how the agent operates (like the body of a SKILL.md file)

Plus Virtual MCP capabilities:

• Name - Identifies the agent (e.g., “Deployment Agent”, “Research Agent”)
• Tools - Actions the agent can take (select from connections or define inline)
• Resources - Additional context and documentation
• Prompts - Workflow templates

decopilot uses the description field to understand when to spawn each agent—just like it uses skill descriptions to know when to invoke skills. This makes agents discoverable and automatically available when relevant to the task at hand.

Since agents are Virtual MCPs, you can compose them from existing connections, define capabilities inline, or mix both approaches.

Example: Order Fulfillment Agent

An order fulfillment agent might aggregate Shopify tools for order management, ShipStation tools for shipping label generation, and Slack tools for fulfillment team notifications. The agent handles order fulfillment with a focused toolset—fetch pending orders, check inventory, generate shipping labels, send tracking notifications—all within its single-purpose scope.

Built-in Agents

deco CMS includes built-in agents for common mesh operations. These agents use the Management MCP—a self-referential MCP that exposes mesh management capabilities through the MCP protocol.

Built-in agents available out-of-the-box:

• Agent Management - Create, edit, and delete agents
• Connection Management - Manage MCP server connections
• Permission Management - Control access and authorization
• MCP Store - Browse and install MCPs from the store
• User Management - Invite and manage organization members

These agents operate at organization level and leverage Management MCP tools for platform configuration.

Agent Design Principles

Single-purpose focus: Each agent should do one thing well. Narrow scope improves clarity, reliability, and maintainability.

Clear description: Write descriptions that help decopilot understand when to spawn the agent. Use [Verb] + [Object] + [Context] pattern:

• “Fulfill customer orders and generate shipping labels”
• “Analyze sales data for inventory optimization”
• “Process customer refunds and returns”

Detailed instructions: Provide clear guidelines in the instructions field—just like you would in a SKILL.md file. Include constraints, workflow steps, and behavioral guidelines.

Tool selection: Choose tools that align with the agent’s description. A customer service agent needs order lookup and refund tools; an inventory agent needs stock checking and alert tools. Overly broad tool access dilutes focus.

Good: “Order Fulfillment Agent” (processes and ships orders) Too broad: “Ecommerce Operations Agent” (fulfillment, customer service, inventory, marketing, analytics…)

Using Agents

Agents can be used in several ways:

• Spawned by decopilot - Agents created on-demand with fresh context for specific tasks
• Built-in agents - Pre-configured agents for mesh operations, ready immediately
• Custom agents - Domain-specific agents for your team’s unique workflows

When working with an agent, you’ll only see the tools, resources, and prompts configured for that agent’s specific purpose. This focused view helps you stay on task without distraction.

Benefits Summary

Agents provide focused capabilities with clear advantages:

• Focused capabilities - Single-purpose design ensures clarity and reliability
• Fresh context - Spawned agents start with clean context for their task
• Reusable patterns - Define once, use across projects and workflows
• Built-in operations - Pre-configured agents for common mesh tasks
• Self-referential management - Management MCP enables agents to manage mesh itself

Next steps: Learn about Virtual MCPs to understand the foundation, or explore decopilot to see how agents are spawned dynamically.