AgentBoard plugin

AgentBoard adds an AI sidebar to the browser with your own models and keys, WebMCP tools scripted per page and remote MCP servers.

  • 136 ★ on the repository
  • Last commit
  • MIT
Plugin

Broad access

Why this level

  • The extension reads page content, including the DOM, and sends it to the chosen model with the URL.
  • User WebMCP scripts run with page-level capabilities and may act inside signed-in accounts.
  • Model keys and MCP tokens sit in settings, and exports store them in plaintext.
All reasons and checks

igrigorik/agentboard

Install

Manual install

Text for your agent

Install the extension from the Chrome Web Store link in the README and create an agent profile with an explicit connection API and your own key. Test `agentboard_read_page` on a safe page, then add your own WebMCP scripts.

Access and checks

How we review

Broad accessThe tool can change data, run commands or handle secrets.

  • The extension reads page content, including the DOM, and sends it to the chosen model with the URL.
  • User WebMCP scripts run with page-level capabilities and may act inside signed-in accounts.
  • Model keys and MCP tokens sit in settings, and exports store them in plaintext.

This is third-party code. Review the repository files before installing.

What it does

AgentBoard is a Chromium extension that opens a tab-scoped AI sidebar. You can create several agent profiles and switch between them mid-conversation. The connection is chosen explicitly: OpenAI Responses or Chat Completions, Anthropic Messages, Google Generative AI or a compatible proxy, including local Ollama. On pages AgentBoard registers tools through `document.modelContext.registerTool()`, so user WebMCP scripts work like Greasemonkey for AI. Remote MCP servers, templated commands such as `/analyze` and built-in system tools, for example reading a page or PDF, are added on top. A local folder can be connected for agent behavior and durable memory.

Who it is for. Power users of the browser and WebMCP developers who want their own model in a sidebar and tool scripts for specific sites.

Good fit when

  • A model on your own key or a local one is needed next to a page without a third-party proxy.
  • A WebMCP tool that clicks and reads the DOM is written for a site.
  • Remote MCP servers with APIs and databases are connected to a browser AI.
  • A page or PDF is analyzed as bounded Markdown with a screenshot.
  • Frequent requests are reduced to command templates with arguments.

Not a fit when

  • Pages hold private data that must not go to the chosen model.
  • You need an agent for a terminal and repository rather than a browser tab.
  • Page scripts would run without your review.

Example request

Read the open page, pull out the key figures and call my WebMCP tool to save the table.

Limitations

The extension runs in Chromium, installs from the Chrome Web Store and builds from source into `dist/`. The conversation goes to the chosen model endpoint with page URL and title snapshots, including local `file://` paths. User WebMCP scripts run with page-level capabilities. The built-in YouTube transcript tool may send browser session credentials. Settings exports hold keys and tokens in plaintext.

How to disable. Disable or remove the extension at `chrome://extensions`. First delete profiles with keys and MCP tokens in settings, and erase any settings exports.

FAQ

Where do I get the extension?

From the Chrome Web Store via the README link. For development the repository builds into `dist/`, which you load through `chrome://extensions` in developer mode.

Which models are supported?

Any model through four connection types: OpenAI Responses, OpenAI Chat Completions, Anthropic Messages and Google Generative AI. Local models such as Ollama go through a compatible endpoint.

What is a WebMCP tool?

A script that registers a function through `document.modelContext.registerTool()` on a page. The agent calls it in the tab context.

Does AgentBoard have its own telemetry?

No. Settings live in `chrome.storage.local`, and requests go to the model, MCP and website addresses you configured.

Why is a settings export risky?

The file holds keys, MCP tokens, endpoints and user scripts in plaintext. Treat it as a secret.

Discussion

No questions or reviews yet.

Plugins

An MCP bridge to control your browser from a coding agent: a Chrome extension and a set of tab and page automation tools

MCP serverBroad access1.3KRepository stars

A browser extension and local proxy that bring MCP tools into web chat UIs like ChatGPT, Perplexity and Grok

CLIExtended access2.6KRepository stars
Official

A remote MCP server and Chrome extension: the agent reads data behind your logins and operates the sites you already use

MCP serverBroad access1.6KRepository stars
Official

An MCP server and Chrome extension that let the agent drive your already open browser with your logged-in sessions, all locally

MCP serverBroad access7.3KRepository stars

A Chrome extension with an MCP server: the agent drives your everyday browser with its sessions, tabs, bookmarks and history

MCP serverBroad access12.5KRepository stars

AgentKey plugin links an agent to a paid cloud data gateway: web search, social media, quotes and on-chain data in one plan

PluginExtended access655Repository stars

Comparison with similar tools

Cards that share the main search query, and how they differ.

ToolOfficialAuthenticationRepository starsMain difference
AgentBoard (this card)noAPI key136
Browser MCPyesnot required7,345hands browser control to your external client; AgentBoard opens its own AI panel in the tab and connects your models
OpenTiny WebMCP SDKno125an SDK for publishing WebMCP tools on a site; AgentBoard calls them and lets users script tools for other people's pages

README in short

Privacy is documented in a separate `PRIVACY.md`: what goes to which endpoint and how to delete data. The extension keeps conversations in memory, not in extension storage.

README badge

Report a problem