> ## Documentation Index
> Fetch the complete documentation index at: https://docs.cruq.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Agent Builder

> Describe the agent you need in chat. The Agent Builder designs it, builds it, tests it and hands it over.

The Agent Builder is a team of three agents that build other agents for you.
You describe what you need in plain words, approve the design, and get a
published, tested agent back. No clicking through the builder.

| Agent | What it does |
| - | - |
| **Agent Builder** | The one you chat with. Asks what you need, shows you the design, builds it once you say yes, and tells you what to do next. |
| **Agent Designer** | Turns your request into a full spec: model, instructions, tools, starter prompts and test cases. It only uses what your workspace actually has. |
| **Agent Tester** | Runs the new agent against the test cases and reports what passed and what failed. The Builder fixes clear failures and tests once more. |

## Set it up

The team works through the **Cruq** integration, which gives it your
workspace's [MCP server](/docs/mcp) with build tools.

1. Under **Settings > API keys**, create a key with the **Management** type.
   It starts with `cruq_mk_`.
2. Under **Integrations**, connect **Cruq** and paste that key. It is stored
   encrypted, like every other integration secret.
3. Open **Agent Builder** in the playground and describe the agent you want.

The Agent Builder team is added to a workspace by the Cruq team. If you don't
see it under Agents, ask us at [hello@cruq.ai](mailto:hello@cruq.ai).

## A conversation

> I run a bakery in Dubai. I want an agent that answers questions about our
> menu and opening hours, and takes custom cake orders: name, phone, date,
> flavour and size. English and Arabic.

The Builder asks anything it needs to know, then shows you the design: the
agent's name, what it does, the model, its tools, a summary of its
instructions, and anything you need to connect or upload. It builds nothing
until you say yes.

After building, it tests the agent and hands it over with next steps: try it in
the playground, upload your menu to the knowledge base, or connect it to
WhatsApp.

## What it can and can't do

It can:

* Create agents and change existing ones. Each change is published as a new
  version, so you can roll back from the agent's history.
* Pick builtin tools (knowledge search, memory, ask a human, browser on the
  sites you name, and others) and integrations you've already connected.
* Make one agent delegate to another.
* Create knowledge bases and add text you paste into the chat.

It can't:

* Delete anything. Deleting agents and knowledge bases stays in the dashboard.
* Connect integrations or handle keys. You connect those yourself under
  **Integrations**, and the Builder tells you which ones would help.
* Read files you attach. Upload documents under **Knowledge**.
* Edit agents with routers or several agent nodes, which are built on the
  canvas. It tells you to edit those in the dashboard.

## Use the build tools from your own client

The same tools are on the MCP server for any management key, so Claude,
Cursor or another MCP client can build Cruq agents too. See
[MCP server](/docs/mcp#build-tools).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.