> ## 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

> Connect AI agents to 2extract: provision proxies over MCP, feed our documentation to any model, and search the docs from your editor.

2extract is built to be driven by software, not only by people. If you work with AI agents - Claude Code, Cursor, Codex, or something you wrote yourself - there are three ways to plug 2extract into them.

<CardGroup cols={3}>
  <Card title="MCP Server" icon="plug" href="/ai/mcp-server/overview">
    Let an agent create proxies, resolve geo targeting, set spend limits, and read usage.
  </Card>

  <Card title="LLM References" icon="file-lines" href="/ai/for-agents/llm-references">
    Our whole documentation portal as two machine-readable text files, following the llms.txt standard.
  </Card>

  <Card title="Docs MCP Server" icon="magnifying-glass" href="/ai/for-agents/docs-mcp-server">
    Let an MCP client search this documentation live, instead of loading it into context.
  </Card>
</CardGroup>

***

## Which one do you need?

The three are complementary, and most teams end up using more than one.

| You want the agent to...                                | Use                                               |
| :------------------------------------------------------ | :------------------------------------------------ |
| Create and configure proxies, check balance, set limits | [MCP Server](/ai/mcp-server/overview)             |
| Know how 2extract works while building your tools       | [LLM References](/ai/for-agents/llm-references)   |
| Look up an answer in our docs on demand                 | [Docs MCP Server](/ai/for-agents/docs-mcp-server) |

The distinction that matters: the **MCP Server acts** - it changes things in your account and costs money. The other two only **read** - they are documentation in a shape a model can consume.

***

## What the MCP Server does

Instead of clicking through the dashboard, you describe the job:

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

The agent resolves the countries, picks a plan, checks your balance, creates a proxy named after the task, sets a traffic cap, and hands back connection strings for each country - ready to paste into your tools.

<CardGroup cols={2}>
  <Card title="Quick Start" icon="rocket" href="/ai/mcp-server/quick-start">
    Connect your first client in about two minutes.
  </Card>

  <Card title="Usage Examples" icon="terminal" href="/ai/mcp-server/usage-examples">
    Full sessions, from the request to the connection string.
  </Card>
</CardGroup>

***

## Supported clients

Any client that speaks the Model Context Protocol over Streamable HTTP works. We keep setup guides for the most common ones:

<CardGroup cols={4}>
  <Card title="Claude Code" href="/ai/clients/claude-code" />

  <Card title="Claude Desktop" href="/ai/clients/claude-desktop" />

  <Card title="Codex" href="/ai/clients/codex" />

  <Card title="Cursor" href="/ai/clients/cursor" />

  <Card title="VS Code" href="/ai/clients/vs-code" />

  <Card title="Windsurf" href="/ai/clients/windsurf" />

  <Card title="Cline" href="/ai/clients/cline" />

  <Card title="Other clients" href="/ai/clients/other" />
</CardGroup>

***

## What it costs

The MCP server itself is free. You pay for the proxies your agent creates, at the same pay-as-you-go rates as everything else - see [Billing & Usage](/guides/billing-usage).

<Warning>
  An agent with a `proxies:write` key can create billable resources on your account. Scope the API key to what the agent actually needs, and set a spend limit on every proxy. See [Provisioning & Cost Control](/ai/use-cases/provisioning-cost-control).
</Warning>