---
title: Getting Started | cadenya
description: Create your first tool set, configure an agent, and run an objective.
---

This guide walks you through the core Cadenya workflow: creating a tool set, configuring an agent, and running an objective. By the end, you’ll have an agent executing a task using tools you’ve defined.

You’ll need a Cadenya account, a workspace, and an API key. See the [API Reference](/api/index.md) for authentication details.

## Create a tool set

Tool sets connect Cadenya to external tool providers. In this example, we’ll create a tool set that syncs tools from an MCP server.

1. ### Define the tool set

   Create a tool set with an MCP adapter. The adapter tells Cadenya where to find tools and how to connect.

   ```
   const toolSet = await client.toolSets.create({
     name: "My MCP Tools",
     description: "Tools synced from my MCP server",
     adapter: {
       mcp: {
         url: "https://your-mcp-server.example.com",
       },
     },
   });
   ```

   Cadenya connects to the MCP server and syncs the available tools into the tool set. You can view the synced tools by listing the tool set’s tools.

## Create an agent

Agents define behavior through variations. Each variation specifies a system prompt, a model, and which tools are available.

1. ### Create the agent with a variation

   Create an agent and include an initial variation that uses the tool set from the previous step.

   ```
   const agent = await client.agents.create({
     name: "My First Agent",
     description: "An agent that uses my MCP tools",
     status: "published",
     default_variation: {
       name: "v1",
       prompt:
         "You are a helpful assistant. Use the tools available to you to complete the user's request.",
       model_config: {
         model_id: "claude/opus-4.6",
         temperature: 0.5,
       },
       agent_tools: [
         {
           tool_set: {
             tool_set_id: toolSet.metadata.id,
           },
         },
       ],
     },
   });
   ```

   Setting `status` to `"published"` makes the agent immediately available for objectives. You can also create agents in `"draft"` status and publish them later.

## Run an objective

Objectives are where agents do work. You provide an initial message describing the task, and the agent executes it.

1. ### Create the objective

   ```
   const objective = await client.objectives.create({
     agent_id: agent.metadata.id,
     initial_message: "Summarize the latest data from my dashboard.",
   });
   ```

   The objective starts in a `pending` state, transitions to `running` as the agent begins work, and eventually reaches `completed` or `failed`.

2. ### Check the result

   Retrieve the objective to see its current status and execution details.

   ```
   const result = await client.objectives.retrieve(objective.metadata.id);


   console.log(result.status.state); // "completed"
   ```

   You can also list events and tool calls to see exactly what the agent did during execution.

## What’s next

Now that you’ve seen the full workflow, explore each concept in depth:

- [Tool Sets](/guides/tool-sets/index.md) — Adapters, tool management, and filtering
- [Agents](/guides/agents/index.md) — Variations, lifecycle, and tool assignment
- [Objectives](/guides/objectives/index.md) — Monitoring, continuing, and providing feedback