The pages of Invantive Cloud Documentation offer their search and their content as tools to an AI agent that runs in the browser of the reader. They follow the [WebMCP proposal](https://github.com/webmachinelearning/webmcp) of the Web Machine Learning community group of the W3C: a page registers tools through `navigator.modelContext`, and the agent of the browser calls them instead of reading the screen. ## Tools Every page registers the following tools: - `search` (`query`, optional `language`): searches the documentation, the data models and the forum answers, and returns the hits as Markdown. - `getPage` (`address`): reads a page of the site as Markdown, given its address such as `/en/docs/...`. - `listTables` (`provider`): lists the tables of the data model of a platform, given the provider part of the address such as `exactonline`. - `getTable` (`provider`, `table`): reads the columns of a table, given the provider part of the address and the full table name `catalog.schema.table`. The tools return the same Markdown as [[Invantive Cloud Documentation/Interfaces/Markdown for AI Agents|Markdown for AI Agents]] and answer in the language of the page unless a language is given. ## Behaviour The tools are registered only in a browser that offers `navigator.modelContext`; in any other browser the script does nothing. They need no consent, since they set no cookie and contact no other party, and they expose nothing that the page does not show. They are subject to the same rate limits as the pages. The WebMCP proposal is not yet a standard. The names and inputs of the tools may change when the proposal changes.