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

# AgentWeb documentation

> AgentWeb documentation for connecting an AI agent, discovering supported website actions, authenticating, executing, and verifying results.

AgentWeb connects an existing AI agent to supported web actions through MCP (Model Context Protocol) and REST API. Every live tool, including `read_page` and `discover`, requires an AgentWeb account and `aw_` API key; account creation runs in-conversation through the one-link sign-in flow. Check the returned status or verification before reporting completion.

## Does it work on my site?

AgentWeb currently maps over 1,100 sites, each with named, executable actions. Call `list_actions` for the live catalogue (paginated, with per-site action counts), or browse the [mapped-site library](https://www.agentweb.us/sites). Passing an exact domain to `list_actions` returns that site's full action definitions with parameters. If a site is not mapped yet, `request_map` queues a build and `agentweb_map_status` reports progress.

## Entry point

* **Use AgentWeb**: connect the MCP server and run mapped website actions for a user. Start with the [quickstart](https://docs.agentweb.us/quickstart.md).

## Start here

* [AgentWeb quickstart](https://docs.agentweb.us/quickstart.md)
* [MCP server and tool definitions](https://docs.agentweb.us/mcp-apps.md)
* [Authentication and target-site connections](https://docs.agentweb.us/agent-auth.md)
* [Developer overview](https://www.agentweb.us/developers)
* [API reference](https://www.agentweb.us/api-reference)
* [Public MCP tool manifest](https://www.agentweb.us/.well-known/mcp/tools.json)
* [Full agent-readable reference](https://www.agentweb.us/llms-full.txt)

## Execution sequence

1. Authenticate with an `aw_` API key (create one with the one-link sign-in tools).
2. Read a page or call discovery for the target domain, then list mapped actions.
3. Connect the target-site account if the selected action requires it.
4. Confirm the inputs and obtain approval for consequential actions.
5. Execute once and use the structured result to report success, failure, or required intervention.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.