import { CardGrid, LinkCard } from '@astrojs/starlight/components';
import Faq from '@components/Faq.astro';

**Aymar Agents** has its own **MCP** server ([Model Context Protocol](https://modelcontextprotocol.io)): it
exposes your account — leads, contacts, conversations, appointments and tickets — as a set of *tools* your AI
assistant can read and, with your confirmation, write to. It's exactly what the public API allows, not one
endpoint more — its tools are generated from the same OpenAPI definition.

<CardGrid>
  <LinkCard title="What it is and what it is not" href="/en/mcp/getting-started/" description="A deterministic adapter over your own public API, with no language model of its own." />
  <LinkCard title="Connect your assistant" href="/en/mcp/connect/" description="URL, X-Api-Key header and Streamable HTTP transport." />
</CardGrid>

## Connect your client

<CardGrid>
  <LinkCard title="Claude Desktop" href="/en/mcp/clients/claude-desktop/" description="Desktop client configuration." />
  <LinkCard title="Claude Code" href="/en/mcp/clients/claude-code/" description="Via CLI or with .mcp.json." />
  <LinkCard title="Cursor" href="/en/mcp/clients/cursor/" description="Editor configuration." />
</CardGrid>

## Reference

<CardGrid>
  <LinkCard title="Tool catalog" href="/en/mcp/tools/" description="Every available tool, generated from the same OpenAPI definition as the public API." />
  <LinkCard title="Confirmation on writes" href="/en/mcp/confirm/" description="No write happens without your explicit confirmation." />
  <LinkCard title="Limits, errors and troubleshooting" href="/en/mcp/troubleshooting/" description="What to do when something fails." />
</CardGrid>

## Related tools

<CardGrid>
  <LinkCard title="API reference" href="/en/api/" description="The same surface, in plain REST." />
  <LinkCard title="llms.txt" href="/llms.txt" description="An index of the whole portal in plain text." />
</CardGrid>

## Frequently asked questions

<Faq items={[
  { question: 'How is the own MCP server different from an external MCP server (BYO-MCP)?', answer: "They run in opposite directions: Aymar Agents's MCP server exposes your account to your external AI assistant (Claude Desktop, Claude Code, Cursor); an external MCP server plugs third-party tools into the agent that talks to your own customers, and it's connected from Settings, not from your MCP client. See <a href=\"/en/mcp/getting-started/\">What it is and what it is not</a>." },
  { question: 'What do I need to connect my MCP client?', answer: "A Pro or Enterprise plan and an API key with the scopes of the resources you want to use — the same mechanism you'd use to call /public/v1 directly." },
  { question: 'Can the MCP server do more than the public API?', answer: 'No — its tools are generated by reading the exact same OpenAPI definition as /public/v1. Anything you can do through MCP you could also do with a direct curl call to the API.' },
  { question: 'Why does my assistant ask for confirmation before creating or changing something?', answer: 'Every write tool requires the confirm: true argument; without it, the server rejects the call before touching the API and returns a summary of what it was about to do. It\'s the same "the LLM is never the only barrier" principle that governs the rest of the product. See <a href="/en/mcp/confirm/">Confirmation on writes</a>.' },
  { question: 'Can an MCP tool delete data?', answer: 'No — no tool can delete anything, because the public API has no DELETE endpoint at all.' },
  { question: 'I only see the status tool in my MCP client, nothing else. Why?', answer: "Either your tenant isn't on a Pro/Enterprise plan, or your key has no read scope at all. Call the status tool to confirm which one it is. See <a href=\"/en/mcp/troubleshooting/\">Limits, errors and troubleshooting</a>." },
]} />

## Up next

<LinkCard title="Connect your assistant" href="/en/mcp/connect/" description="Configuration blocks for Claude Desktop, Claude Code and Cursor." />