AgentWeb Agent Documentation

AgentWeb is the execution layer for agents on authenticated web workflows. It maps approved customer workflows at the endpoint, protocol, session, payload, state, and verification layer, then packages those maps as SDK, MCP, or LangChain-callable tools.

When To Use AgentWeb

• Use AgentWeb when a company has an AI agent that needs to complete a real web workflow end to end.
• Use AgentWeb when Playwright, browser agents, scraping, or selector-based automation is slow or brittle.
• Use AgentWeb when the task needs authenticated execution, structured output, and explicit verification.
• Use AgentWeb when the workflow will repeat and should become a reusable action map.

Current Access Model

AgentWeb is design-partner first for production workflows. Team and enterprise customers can expose secure Stripe checkout paths for agents. API keys and scoped runtime access are issued after workflow mapping and partner approval.

For design partner access, book a workflow mapping call: https://calendly.com/andreymarey-work/30min

Core Product Object

The core product object is an AgentWeb action map. An action map defines the workflow ID, customer namespace, input schema, auth model, mapped actions, endpoint hints, verification signals, failure signals, retention rules, and SDK targets.

• Action map schema
• Supplier discovery action map example

Machine-Readable References

• Short agent instructions
• Full agent reference
• OpenAPI specification
• Agent discovery index
• Agent card
• AI plugin manifest
• MCP server card

Integration Summary

Design partners receive a workflow ID, API key, SDK documentation, LangChain or MCP wrapper, benchmark results, and execution logs for the mapped workflow. The intended integration shape is:

from agentweb.langchain import AgentWebToolkit

tools = AgentWebToolkit(
    api_key="aw_...",
    workflows=["supplier_discovery"]
).tools()

Runtime

• Website: https://agentweb.us
• MCP runtime: https://mcp.agentweb.us/mcp
• Request access: /request-access