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

# AI Assistants

> Connect Relayer docs to Claude Code, Cursor, ChatGPT, and any MCP-compatible tool

The Relayer documentation ships with first-class AI integration. You don't need to copy/paste pages into a chat anymore — your AI assistant can query the docs directly as a tool.

<Info>
  Built into Mintlify: an MCP server, contextual menus, and `llms.txt` — all already live for this site.
</Info>

## Why connect via MCP

The **Model Context Protocol** lets AI clients (Claude Code, Cursor, ChatGPT, Windsurf, Devin, VS Code) search and read this documentation **as a tool call** during a conversation. Concretely:

* Your AI assistant has up-to-date answers about every endpoint, every parameter, every flow — not outdated training data
* The assistant decides when to consult the docs; you don't have to attach pages manually
* Auth, search, and full-page retrieval all work out of the box

## The MCP server URL

```
https://docs.relayer.fi/mcp
```

This URL exposes two tools to any connected client:

* **search** — finds relevant snippets across all docs, returns titles + links
* **query docs filesystem** — retrieves full page content with shell-style commands

## Connect to Claude Code

Run this in your terminal:

```bash theme={null}
claude mcp add --transport http relayer-docs https://docs.relayer.fi/mcp
```

That's it. New Claude Code sessions will automatically discover and use the Relayer docs as a tool.

Verify it's connected:

```bash theme={null}
claude mcp list
```

## Connect to Cursor

Add this to your `~/.cursor/mcp.json` (or the project's `.cursor/mcp.json` for per-repo scope):

```json theme={null}
{
  "mcpServers": {
    "relayer-docs": {
      "url": "https://docs.relayer.fi/mcp"
    }
  }
}
```

Restart Cursor. The MCP server appears in Settings → MCP.

## Connect to VS Code

VS Code's MCP integration is configured via the same `mcp.json` format inside your workspace:

```json theme={null}
{
  "mcpServers": {
    "relayer-docs": {
      "url": "https://docs.relayer.fi/mcp"
    }
  }
}
```

Or use the **Connect to VS Code** button in the contextual menu on any page — it opens VS Code with the server preinstalled.

## Connect to ChatGPT / Claude / Perplexity / Grok

These web-based assistants don't use MCP directly. Instead, the documentation gives them context **per page** through the contextual menu:

<Steps>
  <Step title="Open any page in the docs">
    Navigate to the page you want to ask about.
  </Step>

  <Step title="Click the contextual button (top right)">
    You'll see options for **Copy as Markdown**, **View as Markdown**, and direct deeplinks to ChatGPT / Claude / Perplexity / Grok / Google AI Studio.
  </Step>

  <Step title="Pick a target">
    The chosen tool opens with the current page already attached as context. Ask your question — the assistant has the page content available immediately.
  </Step>
</Steps>

## Use `llms.txt`

Every Mintlify-hosted docs site exposes a curated index of LLM-friendly Markdown at:

```
https://docs.relayer.fi/llms.txt
```

Feed this URL to any LLM workflow that wants a complete machine-readable summary of the documentation tree. It's an indexed table of contents with every page link in plain text — ideal for retrieval-augmented generation pipelines or one-shot doc ingestion.

For full page content, individual pages are available as Markdown at `<page-url>.md`:

```
https://docs.relayer.fi/api-reference/introduction.md
https://docs.relayer.fi/signing/flow-guide.md
```

## What the assistant can answer well

Once connected, your AI tool can answer questions like:

* *"How do I create an agent with a \$500 monthly USDC budget?"* — it queries `/agent/flow-guide` and `POST /v1/agents/{id}/budget`
* *"What headers does the agent SDK send?"* — it pulls `get-started/authentication`
* *"Generate a Node.js snippet to prepare and confirm a Solana transaction"* — it combines the signing flow guide with the endpoint params
* *"Which endpoints in Payout are public vs dashboard-only?"* — it consults the API Reference tags and the architecture page

## What it can't answer (yet)

The MCP server returns what's in the docs. It doesn't have access to:

* Your workspace's specific API keys, agents, or wallet IDs (use the dashboard for those)
* Live data from your account (balances, orders, etc.) — for that, the assistant would need to call the API itself with your credentials
* Information not published in this site (internal roadmap, pricing for enterprise tiers, etc.)

## Get help

* **MCP issues**: try `claude mcp remove relayer-docs` and re-add. If problems persist, [open a support email](mailto:dev@relayer.fi).
* **AI assistants giving wrong answers**: this likely means the docs themselves are unclear or stale — please report at [dev@relayer.fi](mailto:dev@relayer.fi) and we'll fix the page.

## Next Steps

<CardGroup cols={2}>
  <Card title="Quickstart" icon="rocket" href="/get-started/quickstart">
    Make your first API call in under 5 minutes.
  </Card>

  <Card title="API Reference" icon="terminal" href="/api-reference/introduction">
    Browse the full endpoint surface with an interactive playground.
  </Card>
</CardGroup>