Manifest

Manifest is a tool designed to convert any webpage into a structured JSON map that AI agents can utilize for actions such as clicking, filling, and submitting information. It is built for developers and anyone engaged in creating browser-based AI agents, aiming to simplify the interaction between AI and web interfaces. The core problem Manifest addresses is the fragility of traditional web automation methods, particularly the reliance on CSS selectors that break easily with website updates. Existing tools often provide browser control or content extraction, but lack a comprehensive understanding of interactive elements and their dependencies. This gap makes it difficult for AI agents to reliably navigate and perform tasks on dynamic web pages, leading to frequent maintenance and debugging. One of Manifest's key features is its ability to generate a structured JSON action manifest from any webpage. This manifest details what an AI agent can interact with, including clickable elements, fillable forms, and submit buttons. It achieves this through a single API call, bypassing the need for manual and often brittle selector writing. Another significant capability is the inclusion of resolved CSS and role locators for each action. This ensures that the identified elements are robust and less prone to breaking when the website's structure changes. By providing these stable locators, Manifest enhances the reliability of AI agent interactions. A crucial feature is the `requires` field, which encodes dependencies between elements. For example, it can specify that a submit button is only clickable after a particular plan has been selected. This dependency mapping provides essential context that traditional accessibility snapshots or simple element lists do not offer, enabling more intelligent and sequential agent behavior. Manifest offers a Python SDK, integration with LangChain, and an MCP server, making it accessible and easy to incorporate into existing development workflows. These components are designed to streamline the process of building and deploying browser agents. The product works by taking a webpage as input and processing it to create a detailed action manifest. This process involves using tools like Playwright for browser interaction and DOM analysis, combined with an LLM (specifically Claude Sonnet) to infer relationships and dependencies between interactive elements. The output is a JSON object that clearly outlines actionable steps and their prerequisites. The primary benefit for users is the creation of more reliable and robust browser agents. By eliminating fragile selectors and providing explicit dependency information, Manifest reduces the debugging overhead and maintenance costs associated with web automation. This allows developers to focus on the AI's logic rather than the intricacies of web scraping and interaction. Manifest can be used in various scenarios, such as automating form submissions, navigating complex multi-step processes like checkouts, or enabling AI agents to interact with dynamic web applications. For instance, an agent could use Manifest to reliably fill out a signup form, select a service plan, and then submit the request, all guided by the structured action map. Manifest is built for anyone shipping browser agents and is available as a REST API. It includes a Python SDK and LangChain support. The product is currently free to try with no signup required for its demo. In summary, Manifest provides a robust and context-aware method for AI agents to interact with webpages, overcoming the common challenges of selector fragility and action dependencies to enable more reliable web automation.