> ## 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 MCP server and tool definitions

> AgentWeb MCP server setup, public discovery files, core web action tools, authentication requirements, and safe execution flow.

AgentWeb exposes website reading, action discovery, account setup, target-site connection, and mapped execution through a remote Streamable HTTP MCP server.

## Server and public metadata

* MCP endpoint: [https://mcp.agentweb.us/mcp](https://mcp.agentweb.us/mcp)
* Tool manifest: [/.well-known/mcp/tools.json](https://www.agentweb.us/.well-known/mcp/tools.json)
* Server card: [/.well-known/mcp/server-card.json](https://www.agentweb.us/.well-known/mcp/server-card.json)
* OpenAPI: [/openapi.json](https://www.agentweb.us/openapi.json)
* Full agent reference: [/llms-full.txt](https://www.agentweb.us/llms-full.txt)

## Core execution tools

Every live tool below requires an AgentWeb `aw_` API key.

* `read_page`: retrieve accessible page content as structured text and metadata.
* `discover`: inspect AgentWeb coverage and entry points for a domain.
* `list_actions`: list curated sites and actions; an exact domain returns that site's full action definitions.
* `execute`: the core capability. Reserve one credit, run a mapped action, verify completion, settle only verified work, and return a support receipt. Also available over REST as `POST /api/execute`.
* `request_map` and `agentweb_map_status`: queue a curated action map build for an unmapped site and poll its progress.
* `get_usage`: check remaining credits and whether the preceding calls were charged.

## Account and connection tools

* `agentweb_start_claim_link_signup` and `agentweb_poll_claim_link_signup`: preferred one-link AgentWeb account flow.
* `agentweb_start_api_key_signup` and `agentweb_complete_api_key_signup`: email-code fallback.
* `agentweb_start_connection` and `agentweb_poll_connection`: connect a target-site account when required.
* `agentweb_auth_status` and `agentweb_connection_status`: inspect current AgentWeb and target-site state.
* `agentweb_set_location`: save the user's home location for near-me tasks.
* `agentweb_identify`, `agentweb_feedback`, `agentweb_autonomous_mode`: register an email, send feedback, and record explicit autonomous-mode consent.

## Installation and billing flow

1. `agentweb_scan_business`: inspect a company URL for agent readiness.
2. `agentweb_install_agentweb`: return an Agent Tool install plan, PR files, pricing package, Dodo payment authority requirements, and verification checklist.
3. `agentweb_price_setup`: return setup package and payment requirements.
4. `agentweb_create_checkout_session`: hand the signed-in customer a Dodo hosted checkout link for the Starter plan. Enterprise is sales-led via [https://agentweb.us/demo](https://agentweb.us/demo). Agent-held API keys cannot start billing directly.
5. `agentweb_verify_install`: verify deployed discovery surfaces and return proof.

## Action execution flow

1. Use the claim-link flow and reconnect with the returned `aw_` key (only the signup tools and `agentweb_auth_status` work unauthenticated).
2. Read or discover the target domain.
3. List actions and select one supported contract.
4. Connect the required target-site account, if any.
5. Confirm consequential inputs, execute once, and report the returned state.

The manifest is the current machine-readable source for tool names and authentication requirements. Integrations should inspect it rather than hard-coding an assumed catalog.