Lens MCP
Provides browser-grade web access by rendering pages to markdown, taking screenshots, and fetching images using headless Chromium/Firefox (Playwright) and httpx.
README
Lens MCP
A FastMCP server that gives an MCP client browser-grade web access: render pages to markdown, screenshot them, and fetch images — backed by headless Chromium/Firefox (Playwright) and httpx.
Tools
| Tool | Signature | Returns | Description |
|---|---|---|---|
fetch_page |
url, wait_for_selector=None, timeout_ms=15000 |
markdown str |
Renders the page in headless Chromium, falls back to Firefox on HTTP/2 or connection errors, and converts HTML to markdown with links and images preserved. |
screenshot_page |
url, full_page=False, timeout_ms=15000 |
PNG Image |
Screenshots the page in a 1280×800 viewport. Same Chromium→Firefox fallback. Set full_page=True for the full scrollable page. |
fetch_image |
image_url, referer=None |
Image |
Downloads an image over HTTP with a desktop Chrome User-Agent and optional Referer. Format is inferred from Content-Type. |
Requirements
- Python
>=3.14 - uv
- Playwright browsers (
uv run playwright install chromium firefox)
Configuration
| Variable | Default | Description |
|---|---|---|
LENS_PORT |
8788 |
HTTP port the server listens on. |
LENS_REDIRECT_URL |
(unset) | If set, GET / and any non-MCP URL redirects (302) here. Returns 404 when unset. |
LENS_DB_PATH |
lens.db |
Path to the SQLite request log database. |
Running locally
uv sync
uv run playwright install chromium firefox
uv run python app/server.py
The server listens on http://0.0.0.0:8788.
Running with Docker
docker compose up --build
The image is based on mcr.microsoft.com/playwright/python with Chromium and Firefox preinstalled.
Endpoints
| Path | Description |
|---|---|
/mcp |
MCP HTTP transport endpoint |
/icon.svg |
Server icon |
/get |
302 redirect to LENS_REDIRECT_URL, or 404 |
Request logging
Every tool call is logged to a SQLite database (lens.db by default) via a background queue writer. The requests table stores tool name, arguments, status (ok/error), response preview, error message, and duration in ms.
Resilient navigation
fetch_page and screenshot_page use a two-level fallback:
- Firefox fallback — if Chromium raises
ERR_HTTP2_PROTOCOL_ERROR,ERR_HTTP2, orERR_CONNECTION_RESET, the request is retried once with Firefox. - networkidle timeout — if
networkidletimes out (analytics/websockets keeping the page "not idle"), the current page state is used rather than failing the call. - Clean errors — unhandled navigation failures surface as a single-line
ToolErrorwith thenet::ERR_*code, not Playwright's verbose call log.
Graceful shutdown
The server handles SIGINT (Ctrl-C) and SIGTERM (docker stop) — in-flight requests drain before exit, and the request log writer flushes its queue cleanly.
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。