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

# AI guides

> Pick the AI guide that matches what you're building, from editor setup to conversational execution.

These guides are how-to: wiring a coding agent or editor to yield.xyz for one concrete job. For what AgentKit, the MCP server, and agent skills are and how they fit together, see [AI & Agents](/documentation/ai-agents/index).

<Note>
  Not sure which fits? Most people building an integration want one of the first two: an agent that writes the code (AgentKit builder), or an editor that answers accurately while you write it yourself (editor setup). Reach for AgentKit for Claude Code only once you want to run yields conversationally rather than ship code.
</Note>

## Pick your task

| You want to... | Guide |
| - | - |
| Have the agent write your integration | [AgentKit builder](/guides/ai/agentkit-builder) |
| Have your editor or agent answer from current docs and the live API spec | [Set up your editor](/guides/ai/cursor-setup) |
| Write better prompts and review what the agent produced | [Prompting patterns](/guides/ai/prompting-patterns) |
| Run yields conversationally, or build an agent that executes transactions | [AgentKit for Claude Code](/documentation/ai-agents/agentkit/claude-code) and the [MCP server](/documentation/ai-agents/mcp-server) |

The first three guides are build-time: they ground a coding agent in accurate docs and specs while you write an integration. Editor setup needs no API key; the builder asks for yours before it writes code. The fourth is runtime: it connects a live agent to real yield data and wallet actions.

<Columns cols={2}>
  <Card title="AgentKit builder" icon="hammer" href="/guides/ai/agentkit-builder">
    Generate a working yield.xyz integration from a description of your app.
  </Card>

  <Card title="Set up your editor" icon="code" href="/guides/ai/cursor-setup">
    Give Cursor, Claude Code, or another MCP client live docs and spec access.
  </Card>

  <Card title="Prompting patterns" icon="message" href="/guides/ai/prompting-patterns">
    Prompts that produce reviewable, correct output, and how to check it.
  </Card>

  <Card title="AgentKit for Claude Code" icon="plug" href="/documentation/ai-agents/agentkit/claude-code">
    Discover yields and build transactions conversationally, inside Claude Code.
  </Card>
</Columns>

## The safety rule

Every transaction an agent produces, whether it's generated code calling the API directly or the MCP server's action tools, comes back unsigned. Inspect it before signing; never sign or broadcast on an agent's say-so alone. yield.xyz never signs user transactions and never holds keys. See [Guardrails](/documentation/ai-agents/agentkit/guardrails) for the full set of rules an agent should follow before it builds or submits anything.


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