Getting Started

This guide walks you through the complete end-to-end flow from setting up backend connections to publishing a live agent that users can interact with in AI Workspace.

Overview

Building and deploying an agent in Agent Studio follows this sequence:

Backend Server → MCP Profile → MCP Instance → Skill Registry → Agent Profile → Publish → AI Workspace

Agent Studio supports two agent types:

• Single Agent: A standalone agent combining a trigger, Memory and Cache, an LLM, tools (via MCP servers), and skills.

• Deep Agent: A main agent connected to one or more sub-agents, each with their own components.

Step 1: Connect Your Backend (MCP Server Setup)

MCP servers expose your backend APIs as tools that agents can call. Choose the path that fits your use case.

Path A - Internal MCP (your own backend API)

1. Go to MCP Server > Backend Servers and select Register Backend.
   • Enter your backend server URL, authentication settings, and API details.
   • For full details, see Backend Server > Register a New Backend
2. Go to MCP Profiles and select Create MCP Profile.
   • Map the backend APIs you want to expose as MCP tools.
   • Run Playground to verify the API calls return expected results.
   • Select Save to finalize the profile.
3. Select Provision on the MCP Profile.
   • The provisioned server now appears in MCP Instances as a running instance.
4. Go to MCP Instances, click (⋮) on your instance, and select Playground to validate the live instance.
5. After validation, click (⋮) again and select Publish.
   • The MCP instance is now available to be attached to agents.

Path B - External MCP (third-party MCP server)

1. Go to MCP Registry and select Publish External MCP.
   • Provide the external server endpoint and authentication details.
   • The external MCP is registered and available for agents to use.

Step 2: Create Skills

Skills are reusable logic blocks (code, prompts, or API calls) that extend what an agent can do.

1. Go to Skill Registry and select New Skill.
2. Define the skill type, input/output schema, and implementation.
3. Select Save. The skill is now available to attach to agents.

Tip: Browse the Skill Registry to reuse skills shared by other users or published publicly - you do not need to build everything from scratch.

Step 3: Create the Agent

With your MCP servers and skills ready, you can now build the agent.

1. Go to AI Agent Studio > Agent Profiles and select Create Agent.
2. Choose your agent type:
   • Create Single Agent - for a standalone, single-purpose agent.
   • Create Deep Agent - for a multi-agent workflow with a main agent and sub-agents.
3. On the design canvas, configure the following:

Component	What to do
Trigger	Select how the agent is invoked (e.g., chat message, scheduler, webhook)
LLM	Choose the language model that powers the agent's reasoning
Memory and Cache	Select a memory/cache profile for conversation context
MCP Servers	Attach the MCP instances you provisioned in Step 1
Skills	Attach the skills you created in Step 2

1. Select Save to store the agent configuration.

For detailed configuration steps, refer to AI Agent Studio > Agent Profiles > Create and Update Single Agent Profile or Create and Update Deep Agent Profile.

Step 4: Test in Playground

Before going live, validate that your agent behaves as expected.

1. Go to AI Agent Studio > Agent Profiles. On the agent profile card, click (⋮) and select Playground.
2. Send test messages and verify:
   • MCP tool calls return correct results.
   • Skills execute as intended.
   • The LLM responds accurately within the agent's scope.
3. Adjust the configuration and re-test until the agent performs correctly.

Step 5: Provision and Publish

1. Select Provision on the agent profile as a deployment step.
2. Once provisioning completes, select Publish to make the agent live.
   • The agent is now available in AI Workspace for users to interact with.

Step 6: Use the Agent in AI Workspace

Once published, users can access and interact with the agent from AI Workspace.

1. Open AI Workspace and go to Home.
2. In the right sidebar, select the published agent from the Agent selector.
3. Optionally add Photos & Files, Connectors, or Skills as needed before sending chat.
4. Type a message in the workspace area and interact with the agent.

What's next: Your agent is live and ready for use. You can refine it at any time - update skills, add MCP tools, or adjust configurations - without needing to republish unless structural changes require it. Refer to the AI Workspace section for details on managing agent conversations and settings.