WebAIAssist Brings Browser-Based AI Tools to the Agentic Web
Discover WebAIAssist, a free Chrome extension connecting local or remote AI models directly to website tools using WebMCP drafts.

The landscape of browser automation and AI interaction is evolving with the introduction of WebAIAssist, an open-source Chrome extension developed by software architect Andrei N. Besleaga. Designed to function as a side panel chat assistant, the tool allows users to interact with websites that expose functional tools through an emerging standard known as WebMCP. Instead of relying on traditional visual automation—which can break when layouts change—WebMCP enables web pages to present structured lists of tools, complete with names, descriptions, and required inputs, allowing an assistant to invoke them directly.
Understanding WebMCP and WebAIAssist
WebMCP operates as a draft standard currently incubated within the W3C Web Machine Learning Community Group, with editor involvement from major tech firms like Google and Microsoft. Through this framework, pages can register specific actions, such as searching catalogs, reading articles, or checking orders. WebAIAssist serves as the client-side interface for these tools. When visitors navigate to a supported page, the extension displays the available tools in Chrome's side panel.
Users maintain strict control over execution. The extension requires explicit user approval before running any tool call, showing the precise arguments and the target site in advance. While read-only tools can be configured to bypass prompts on a per-session basis, any actions flagged for review always require a manual click. Furthermore, the extension respects standard machine-readable exclusion rules, refusing to pull text from pages blocked by robots.txt configurations.
Local Execution and Model Flexibility
Privacy is a core design choice for WebAIAssist. The extension features zero site access by default upon installation, requiring manual permission for each individual domain. Users can run small AI models locally on their hardware, utilizing their graphics card, central processor, or Chrome's built-in Gemini Nano. Supported local configurations include lightweight architectures such as Qwen3, Llama 3.2, and Gemma variants.
Alternatively, users can connect external local servers like Ollama and LM Studio, or route requests to remote OpenAI-compatible services. For developers looking to experiment with heavier workloads, you can try top AI models cheaply through one API at https://apixoai.online. Remote configurations require a one-time user authorization before transmitting page text, prompts, and tool outputs to external hosts.
What it means for developers
For developers exploring the agentic web, WebAIAssist pairs with AgenticSystemCore 1.1, an open specification that transforms Markdown notes into semantic websites and knowledge graphs. Version 1.1 introduces structured link definitions, vocabulary tools, and browser-based SPARQL queries without requiring backend servers. Developers can run local nodes or expose tools via Model Context Protocol (MCP) servers for desktop environments.
Testing small models reveals practical engineering constraints. Tests conducted in late 2026 showed that while built-in browser models handled basic tool selection consistently, smaller hardware-accelerated models varied in reliability when navigating live environments. Because WebMCP remains an early draft requiring experimental browser flags in Chrome, developers should treat these implementations as testing grounds while evaluating performance across local and remote setups.
Source: WebAIAssist: One Assistant for the Agentic Web — Towards AI. Written by the Apixo team from that report.
One key for Claude, GPT, GLM, DeepSeek and more. Pay per token with crypto.
Get your API key

