A local AI model in your side panel that uses the WebMCP tools of the open page. Works with AgenticSystemCore nodes.
Users
2
store count
Rating
0.0 / 5
unchanged in 30 days
Ratings
0
total ratings
Store rank
#358K
#295K among extensions
Version
0.1.1
updated 20 Nov 2026
Publisher and listing changes
From our daily record of this listing since 17 Nov 2026
No change of publisher name, contact email, website, privacy policy or listing name since 17 Nov 2026.
About WebAIAssist - Browser AI for WebMCP websites and the agentic knowledge web
WebAIAssist puts an AI inside your browser's side panel, with no server to install, and lets it use the tools a web page offers. By default, before any call runs, you see what it will do and press Allow or Deny.
The AI runs on your computer: on the graphics card, on the processor, or as Chrome's own built-in one. What you type stays in this browser, except the arguments of a tool call you allow, which go to the page that offers the tool, and the chat when you set up a chat server.
Sites can offer tools through WebMCP, a specification that lets a page describe what it can do, such as a search or a form. WebAIAssist lists them with what the page says each does and whether it only reads. By default, for each call you see the name, the site and the exact arguments.
It also reads what a site publishes about itself: llms.txt, a knowledge link set, robots.txt, security.txt and a few other files. The panel shows them. A site that asks for no AI input gets none of its page text or files sent, unless you press Use on this tab anyway (a model in the browser only). For a node of the AgenticSystemCore project, a card shows its version and graph, and you can run read-only SPARQL queries over it.
Before you start:
Chrome keeps WebMCP behind a switch. Open chrome://flags/#enable-webmcp-testing, choose Enabled and relaunch. Without it, sites' tools do not appear.
WebGPU and WebAssembly allow AI models to run locally in user's browser. WebMCP is a draft, not yet a W3C Standard. It may change and not many sites implement it. Needs Chrome or another Chromium browser. The first LLM model start downloads ~ 400 MB to 3 GB, after you agree to use in browser, or you can use a local (eg: Ollama), or remote server model, via an OpenAI-style address (eg: OpenRouter).
Answers can be wrong. Small models often miss the right tool, so check what matters.
No account, no analytics. No access to any site at install; you allow one site at a time.
Works also with AgenticSystemCore, by the same author. Open source, Apache-2.0: https://github.com/andreibesleaga/webaiassist
Version history
2 versions recorded
Updates that changed permissions, site access or size (1 since 17 Nov 2026)