opencode-webmcp
Bridges OpenCode to Chromium's WebMCP API, exposing WebMCP tools from a webpage as standard MCP tools for any MCP client.
README
opencode-webmcp
MCP bridge connecting OpenCode to Chromium's WebMCP API.
Exposes WebMCP tools registered in a Chromium page as MCP tools consumable by any MCP client (OpenCode, Claude Code, etc.).
How It Works
OpenCode (MCP client)
↕ MCP stdio protocol
webmcp-bridge (MCP server)
↕ Puppeteer + CDP
Chromium (WebMCP enabled)
↕ monkey-patched document.modelContext
Web page with registered WebMCP tools
The bridge launches Chromium with --enable-features=WebMCPTesting, navigates to a target page, discovers tools via the WebMCP API, and makes them available to MCP clients.
Monkey-patch: Chrome 150+ has a bug where page.webmcp.invokeTool() returns {} instead of the actual tool result. The bridge works around this by intercepting registerTool to capture execute functions and replacing executeTool to call the captured functions directly.
Prerequisites
- Node.js 18+
- Chromium/Chrome with WebMCP support (Chrome 150+)
- An MCP client (OpenCode, Claude Code, etc.)
Setup
git clone <repo-url> webmcp-bridge
cd webmcp-bridge
npm install
Usage
Standalone
CHROME_PATH=/usr/bin/chromium \
WEBMCP_TARGET_URL="https://www.google.com" \
WEBMCP_HEADLESS=false \
node server.js
OpenCode (opencode.json)
{
"mcp": {
"webmcp-bridge": {
"command": ["node", "/path/to/webmcp-bridge/server.js"],
"enabled": true,
"type": "local",
"environment": {
"CHROME_PATH": "/usr/bin/chromium",
"WEBMCP_TARGET_URL": "about:blank",
"WEBMCP_HEADLESS": "false"
}
}
}
}
Environment Variables
| Variable | Default | Description |
|---|---|---|
CHROME_PATH |
/usr/bin/chromium |
Path to Chromium/Chrome executable |
WEBMCP_TARGET_URL |
https://www.google.com |
Initial page to navigate to |
WEBMCP_HEADLESS |
true |
Run headless (true) or with a visible window (false) |
Bridge-Native Tools
These tools are always available, regardless of what the page exposes:
| Tool | Description |
|---|---|
webmcp_navigate |
Navigate Chrome to a URL and refresh discovered WebMCP tools |
webmcp_status |
Report connection status, current URL, and available tools |
webmcp_evaluate |
Execute arbitrary JavaScript in the page context |
webmcp_invoke_tool |
Call any WebMCP tool by name with arguments |
webmcp_register_test_tools |
Register sample tools (test_greet, test_calculator, test_fetch_title) for validation |
Architecture
server.js— single-file MCP server using@modelcontextprotocol/sdk- Puppeteer (
puppeteer-core) — drives Chromium via Chrome DevTools Protocol - Monkey-patch —
page.evaluateOnNewDocumentinjects a script that interceptsdocument.modelContext.registerToolandexecuteToolto capture actualexecutefunctions
Troubleshooting
- "Failed to parse input arguments" — caused by
document.modelContext.executeTool()failing on the page side. The monkey-patch bypasses this by calling the captured executor directly. - "detached Frame" — the page was navigated away or closed. The bridge auto-recovers by creating a new page. Re-run the tool call.
- No tools discovered — ensure the page has called
document.modelContext.registerTool()with tool definitions. Usewebmcp_evaluateto calldocument.modelContext.getTools()manually. - CORS errors in test tools —
test_fetch_titleusesfetch()which is subject to page CSP. Test on a permissive origin.
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。