> This page is for version v2026-02-09 (default).
> For other versions, use one of these documentation indexes:
> - v2026-02-09 (default): https://docs.extend.ai/2026-02-09/llms.txt
> - v2025-04-21: https://docs.extend.ai/2025-04-21/llms.txt
> - v2024-12-23: https://docs.extend.ai/2024-12-23/llms.txt

> ## Documentation Index
> Fetch the complete documentation index at: https://docs.extend.ai/llms.txt
> Use this file to discover all available pages before exploring further.
>
> ## API version
> The current API version is `2026-02-09`, served at the site root (no version prefix in URLs).
> If this page URL contains `/2025-04-21/` or `/2024-12-23/`, you are reading an older API version.
> Prefer the current docs at https://docs.extend.ai/llms.txt unless the user explicitly needs that older version.
> Do not treat older-version pages as the source of truth for new integrations.

# Agent Quickstart

> Connect an AI agent to Extend three ways: MCP tools for chat clients, the CLI with its installable skill, and a platform context file for agents that write SDK code.

This page helps you equip an AI agent to work with Extend. There are three integration paths, each built for a different kind of agent. Pick the one that matches how yours works.

## Pick how your agent will work with Extend

| If your agent…                                                         | Give it…                                                |
| ---------------------------------------------------------------------- | ------------------------------------------------------- |
| Calls tools from chat (Cursor, Claude, ChatGPT, and other MCP clients) | The [**MCP server**](/mcp)                              |
| Runs one-off and shell-driven document jobs                            | The [**`extend` CLI**](/cli) and its `extend-cli` skill |
| Writes durable integration code with an [SDK](/sdks)                   | The **`extend-api` context file**                       |

The paths are independent. Set up only what your agent will actually use, and add another path later if its role grows. Whichever you pick, your agent can also [read these docs directly](#let-your-agent-read-the-docs) with no setup at all.

> **Note**
>
> The CLI and SDK paths work with any coding agent that can read a skill file or fetch a URL: Claude Code, Cursor, Codex, OpenCode, Goose, and others. The MCP path works with any client that supports remote MCP servers.

## Connect a chat or tool-calling client over MCP

The fastest path, and the only one with nothing to install: no API key, no CLI, no context files. Connect your client to the hosted server and complete the OAuth browser sign-in when prompted:

```text
https://mcp.extend.ai/mcp
```

The server exposes the whole platform (parsing, extraction, classification, splitting, PDF editing, workflows, files, webhooks, and evaluations) as tools with run polling and structured errors built in. During sign-in you choose exactly which workspaces the client may use, and whether it can act in Test, Production, or both.

Per-client setup (Cursor, Claude Code, Claude, ChatGPT, VS Code, Codex, Windsurf), the full tool reference, and troubleshooting live on the [MCP Server page](/mcp).

## Equip an agent that runs the CLI

For agents that do their work in a shell (one-off document jobs, batch runs, and managing extractors, workflows, or webhooks from the terminal), install the CLI and run the setup wizard:

```bash
curl -fsSL https://extend.ai/install.sh | sh
extend setup
```

(Prefer Homebrew, npm, or Go? See the [CLI installation options](/cli#installation).)

The wizard signs you in (through your browser with no API key, or with an API key for headless use), picks your region, and installs the **`extend-cli` agent skill**: a `SKILL.md` generated from the CLI's own command tree that teaches your harness to drive `extend` correctly. It lands at `~/.agents/skills/extend-cli/SKILL.md`, the cross-client default path, with a symlink into `~/.claude/skills/extend-cli` for Claude Code.

For non-interactive environments like CI or a remote agent box, skip the wizard:

```bash
export EXTEND_API_KEY="sk_xxx"   # from https://dashboard.extend.ai/developers
extend skill install
```

> **Tip**
>
> The skill is generated from the installed CLI, so re-run `extend skill install` after upgrading. To scope it to one project instead of your whole machine, pass `--target ./.agents/skills/extend-cli/SKILL.md`.

Verify the setup by asking your agent to run a read-only command that spends no credits:

```bash
extend extractors list
```

A clean list back means the key, the CLI, and the connection all work.

## Equip an agent that writes SDK code

An agent generating application code needs two things: an API key for the code it writes, and Extend's **platform context file** so that code is correct on the first try. The context file is a compact brief on the API, SDKs, schema rules, sync-vs-async patterns, webhooks, and error handling.

First, create a key on the [Developers dashboard](https://dashboard.extend.ai/developers) and export it where the agent's code will run:

```bash
export EXTEND_API_KEY="sk_xxx"
```

SDK clients read the key from the environment (or accept it as a constructor argument) and set the auth header and API version for you.

Then install the context file as the `extend-api` skill from Extend's [agent plugin repository](https://github.com/extend-hq/extend-agent-plugin), which packages it in the [Agent Skills](https://agentskills.io) format every major harness reads:

```bash
npx skills add extend-hq/extend-agent-plugin --skill extend-api
```

This installs at the project level so the configuration can be committed with your repo; pass `-g` to install globally instead. Re-run the same command to refresh the skill after API updates. Prefer zero tooling? The skill is a single file — copy [`skills/extend-api/SKILL.md`](https://github.com/extend-hq/extend-agent-plugin/blob/main/skills/extend-api/SKILL.md) into `.agents/skills/extend-api/` (or `~/.agents/skills/extend-api/` for global use, with a symlink into `~/.claude/skills` for Claude Code).

The official [SDKs](/sdks) cover four languages. Python, TypeScript, and Java ship polling helpers (`create_and_poll` / `createAndPoll`) and built-in webhook signature verification; Go has neither, so an agent writing Go should poll and verify signatures manually.

#### Python

```bash
pip install extend-ai
```

#### TypeScript

```bash
npm install extend-ai
```

#### Java

```text
ai.extend:extend-java-sdk:1.13.0   // Gradle
```

#### Go

```bash
go get github.com/extend-hq/extend-go-sdk
```

## Let your agent read the docs

No setup required. Every page of these docs is directly readable by an agent:

* **Index:** [https://docs.extend.ai/llms.txt](https://docs.extend.ai/llms.txt): a curated map of every doc page.
* **Any page as Markdown:** append `.md` to a doc URL (e.g. `https://docs.extend.ai/extraction/schema.md`) for clean, token-cheap content.

Mention these in your project instructions or add them to your agent's allowed-fetch list so it reaches for the docs instead of guessing. Agents on the MCP path get this for free through the `search_documentation` and `get_documentation` tools.

## Where to go next

#### [MCP Server](/mcp)

Per-client setup, the full tool reference, and troubleshooting.

#### [CLI reference](/cli)

Every command, flag, and output format for driving Extend from the shell.

#### [SDKs](/sdks)

Typed clients, polling helpers, and webhook verification in four languages.

#### [API Quickstart](/api-quickstart)

Call the Extend API directly for more advanced integrations.