> ## Documentation Index
> Fetch the complete documentation index at: https://docs.2extract.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Overview

> What the 2extract MCP server is, what it exposes, and how an AI agent authenticates against it.

The 2extract MCP server exposes your account to an AI agent as a set of callable tools. The agent can create and configure proxies, resolve geo targeting down to city and ASN, cap traffic and spend, and read balance and usage - without you touching the dashboard or writing any code.

It speaks the [Model Context Protocol](https://modelcontextprotocol.io), so any MCP-capable client can connect to it.

***

## Endpoint

```
https://mcp.2extract.com/mcp
```

|                    |                                 |
| :----------------- | :------------------------------ |
| **Transport**      | Streamable HTTP                 |
| **Authentication** | `Authorization: Bearer 2xt_...` |
| **Prompts**        | 4                               |

<Note>
  Keep the `/mcp` path in your client configuration. The bare host serves a landing page, not the protocol.
</Note>

***

## Authentication

Use your regular 2extract API key. There is no separate MCP key type and nothing extra to enable.

```http theme={null}
Authorization: Bearer 2xt_YOUR_API_KEY
```

Create a key in the [Dashboard](https://2extract.com/app/api-keys) under **API Keys**. Full details, including how keys are stored and rotated, are on the [API Keys](/public-api/api-keys) page.

### Scopes an agent needs

Each key carries scopes that gate which tools will work. Grant only what the agent actually needs.

| Scope           | Unlocks                                                                 |
| :-------------- | :---------------------------------------------------------------------- |
| `proxies:read`  | Listing and reading proxy resources                                     |
| `proxies:write` | Creating, updating, and deleting proxies; setting limits and whitelists |
| `balance:read`  | Account balance                                                         |
| `stats:read`    | Traffic and spend statistics                                            |
| `geo:read`      | Country, region, city, ZIP, and ISP lookup                              |

<Warning>
  `proxies:write` does **not** include `proxies:read` - the scopes are independent. An agent almost always reads before it writes, so a write-only key fails partway through most workflows. Grant both.
</Warning>

***

## What the server exposes

MCP servers can offer three kinds of things. This one offers all three.

<AccordionGroup>
  <Accordion title="Tools - what the agent can do" icon="wrench">
    Everything the agent can *do*: account balance and usage, geo lookup, plan lookup, and full proxy lifecycle. Two of them are destructive and ask for your confirmation first.

    See the [Tools reference](/ai/mcp-server/tools).
  </Accordion>

  <Accordion title="Prompts - 4 guided workflows" icon="list-check">
    Ready-made multi-step routines: create a proxy for a stated purpose, build a connection string, report on monthly cost, clean up idle proxies. Clients that support prompts expose them as slash commands.

    See the [Prompts reference](/ai/mcp-server/prompts).
  </Accordion>

  <Accordion title="Resources - what each tool returns" icon="file-code">
    Reference documents the agent reads by itself, so it understands what came back from a tool instead of guessing. Nothing to call and nothing to configure.
  </Accordion>
</AccordionGroup>

***

## A worked example

You ask:

> Collect gaming-laptop prices from Amazon in Germany, the UK and Japan.

The agent, on its own:

<Steps>
  <Step title="Resolves the countries">
    Calls `listGeoCountries` to turn "Germany, the UK and Japan" into values it can target with.
  </Step>

  <Step title="Picks a plan and checks funds">
    Calls `listProxyPlans` and `getAccountBalance` before spending anything.
  </Step>

  <Step title="Creates the proxy">
    Calls `createProxyResource` with a name and description that reflect the task, so the resource is recognisable later in your dashboard.
  </Step>

  <Step title="Caps the spend">
    Calls `setProxyLimits` to put a monthly ceiling on it.
  </Step>

  <Step title="Hands back connection strings">
    One per country, ready to paste into your tools.
  </Step>
</Steps>

The full session, with real output, is on the [Usage Examples](/ai/mcp-server/usage-examples) page.

***

## Next steps

<CardGroup cols={2}>
  <Card title="Quick Start" icon="rocket" href="/ai/mcp-server/quick-start">
    Connect a client and run your first request.
  </Card>

  <Card title="Tools" icon="wrench" href="/ai/mcp-server/tools">
    Everything the agent can reach, and what it cannot do.
  </Card>

  <Card title="Prompts" icon="list-check" href="/ai/mcp-server/prompts">
    The four built-in workflows.
  </Card>

  <Card title="FAQ" icon="circle-question" href="/ai/mcp-server/faq">
    Keys, scopes, limits, errors, and cost.
  </Card>
</CardGroup>