WebMCP tools for agents

WebMCP is an experimental browser API — a W3C Web Machine Learning Community Group draft, not yet shipped by default in any browser — that lets a page declare tools an AI agent can call directly, in your own browser tab, against your own data. It’s the automated version of generating a config with an AI chat tool: instead of you pasting a downloaded config and pasting the result back, an agent that understands WebMCP reads and writes the config already active in your browser, validated the same way either path validates it.

Nothing here changes what the app does for anyone without such a browser or extension — document.modelContext simply won’t exist, and this page’s five tools are never registered. No config is sent anywhere either way: a tool call reads and writes the same localStorage-backed config every other page here already uses.

The five tools

Tool	Does
washy_get_config	Reads the active chart and machine — your own, or the bundled example.
washy_validate_chart	Checks chart rows against the active machine without saving anything.
washy_set_chart	Replaces the active chart, validated first — the same as clicking Save on /config.
washy_set_machine	Replaces the active washer and iron, re-validating the current chart against them — the same as clicking Save on /config/machine.
washy_export_pdf	Renders the active chart as a PDF (phone or print layout), returned as data rather than downloaded.

Every one of them runs through the same @washy-washy/core validation the on-page editors use — a row an agent writes that doesn’t fit your machine fails with the same row-and-column error the chart editor would show you, not a silent guess.

Saved edits reload the page

washy_set_chart and washy_set_machine reload the page once they’ve saved, the same way uploading a config through the header already does — there’s no live syncing between an open editor and a tool call, so a reload is how it catches up.

Trying it yourself

Install a WebMCP-aware browser extension, or open your browser’s devtools console on this site and call await document.modelContext.getTools() — once native support or a polyfill is present, it lists all five tools by name and description before you call any of them.