About AgentWeb

AgentWeb is a web action and execution layer for AI agents. It exposes supported website tasks through maintained Action Maps, MCP, and API so an agent can work with declared inputs, permissions, status, and verification instead of reconstructing every workflow from a screen.

What AgentWeb does

• Reads accessible public pages and discovers mapped actions for a domain.
• Lets an authenticated agent list and execute actions supported by the current map.
• Connects a user's target-site account only when the selected workflow requires it.
• Returns structured completion, failure, and verification information.

Coverage is action-specific, not universal. A mapped domain does not imply that every task on the website is available. Agents should check discovery, permissions, and current status before promising completion.

Official AgentWeb surfaces

• Website: https://www.agentweb.us/
• MCP runtime: https://mcp.agentweb.us/mcp
• Developer guide: /developers
• Product thesis: /thesis
• Founders: About the founders
• GitHub: AgentWeb-Labs/agentweb
• Support: support@agentweb.us

Safety boundary

AgentWeb should not be used for illegal, deceptive, unsafe, unsupported, or unapproved actions. Raw passwords, card numbers, unrestricted credentials, and unrelated private conversation content should not be sent through an action request.