Загрузка видео...
Не удалось загрузить видео
WebMCP is a game changer for browser agents I built a skill that adds WebMCP support to any codebase - in this demo, I turned Excalidraw into an agent-native experience in a few minutes then I used Stagehand to control the canvas and draw anything. works on any open-source codebase
13,605 просмотров • 10 дней назад •via X (Twitter)
Комментарии: 13

skill here:

this is actually interesting because i also think webmcp and opendia with aeon fit together pretty naturally webmcp makes the site agent-readable when the site supports it, but opendia gives any mcp client access to a real browser session when it doesn’t including tabs, cookies, and wallets, with no remote browser or rewrite and aeon is the layer that can keep the workflow running unattended the fastest joint demo would be excalidraw + your add-webmcp skill, controlled from a real chrome tab through opendia and scheduled as an aeon skill try opendia with one command: npx opendia then point a coding agent at the aeon setup skill: or fork the aeon repo and run it locally: cd aeon && ./aeon @aaronjmars is already tying this into webmcp too: if you want to wire your add-webmcp skill into an opendia tab and let aeon schedule it, dm me happy to collaborate.

@Stagehanddev @webmcp challenge : codex draws on tldraw canvas using web-mcp(draw_generetoe) Devpost submit : live website : Video : Repo : Test it in codex browser

@Stagehanddev webmcp still so early

@Stagehanddev so so early

@Stagehanddev does WebMCP expose enough canvas state for @Stagehanddev to recover cleanly when a drawing action partially fails?

@Stagehanddev It should! In this case, /add-webmcp found the Excalidraw API (which is feature complete) and wrapped those methods as tool calls

@Stagehanddev TYPEE SHIT

@Stagehanddev Let’s put code_mode and webmcp into a battle.

@Stagehanddev Frontier evals 💪

@Stagehanddev How does an average Joe end customer use webmcp?

@Stagehanddev How do you decide which canvas actions become typed WebMCP tools versus staying behind browser automation?

@Stagehanddev WebMCP makes the canvas controllable once you’re there. The hole is discovery: can an agent find that it’s callable before it lands. has a grade of 40/100 on AgentBoard.

