Skip to content
Movie Planner

WebMCP tools

WebMCP is a draft browser API — still a nascent W3C incubation, not a ratified standard — that lets a web page declare tools an in-browser AI agent can discover and call, the way this app’s own UI already lets you log, edit, delete, and search your watch history by hand. As of this writing only Chrome has an experimental, flag-gated implementation, so this is progressive enhancement: nothing changes for anyone whose browser or agent doesn’t support it.

Turn it on

Section titled “Turn it on”

These tools are off by default. Turn them on from Settings, next to your other stored credentials: “Let a compatible in-browser agent act on your behalf (WebMCP)”. They’re off by default deliberately — one of the tools below permanently deletes a viewing, and WebMCP’s own consequentialHint (a hint that asks an agent’s UI to confirm before calling a tool marked this way) is a request to the agent, not something this app can enforce. Treat turning this on the same way you’d treat handing someone else the keys to your own calendar app.

What runs, and where

Section titled “What runs, and where”

Exactly the same client-side CalDAV client this app’s own UI already uses (src/lib/caldav/client.ts) — no new server, no new credential store, nothing that leaves your browser. A WebMCP tool’s handler runs in this page’s own JavaScript, the same trust boundary as everything else here: see the README’s “nothing in between” guarantee for the full claim.

The tools

Section titled “The tools”
  • log_viewing — logs a manually watched title (title, when, medium, and the optional fields the manual log form itself offers).
  • edit_viewing — changes fields on an already-logged viewing by its uid, leaving anything not mentioned untouched.
  • delete_viewing — permanently removes a logged viewing. An agent should confirm with you before calling this — it cannot be undone.
  • search_viewings — filters your watch history by the same fields the calendar overview’s own filters use (title, medium, venue, director, actor, genre, city, the movie’s own country/language, rated, and released year/month), capped to a default of 50 matches per call so a large history doesn’t flood the agent’s own context.
  • export_viewings — exports (optionally filtered, same fields as search_viewings) viewings as JSON, the same format Settings’ own “Export as JSON” produces.
  • import_viewings — bulk-imports viewings from CSV or JSON text (the format export_viewings produces, or movie-planner’s own export). A row that looks like a duplicate of something already logged is skipped, never created. A row that matches an existing viewing by uid is only written when the tool is called with mode: "apply-updates" — the default, "create-only", never touches an existing viewing, since there’s no visible per-field review step for an agent’s call the way the real bulk-import page has. There’s no picking some fields and skipping others per row either — an update entry’s changed fields get applied together, so an agent should double-check what it’s about to import before calling this with "apply-updates".