lightpanda-py

Lightpanda for Python, a fast headless browser for AI agents and web automation. Zero dependencies, browser included. The whole thing is one file of about 300 lines: fetch, serve and mcp are the CLI commands, and keyword arguments are the CLI flags.

Installation

pip install lightpanda-py

Example usage

import lightpanda

response = lightpanda.fetch("https://example.com", dump="markdown")
print(response.text)

Output formats

import lightpanda

html = lightpanda.fetch("https://example.com").text
markdown = lightpanda.fetch("https://example.com", dump="markdown").text
data = lightpanda.fetch("https://httpbin.org/ip").json()

# JSON-LD, OpenGraph, Twitter card and meta tags
meta = lightpanda.fetch("https://example.com", dump="structured_data").json()

# Text-only rendering as PNG or PDF
png = lightpanda.fetch("https://example.com", dump="png").content

Waiting and stripping

import lightpanda

# Wait for network idle, or for an element to appear
lightpanda.fetch("https://example.com", wait_until="networkidle")
lightpanda.fetch("https://example.com", wait_selector="h1", dump_selector="h1")

# Drop scripts and styles, or keep only the main content
lightpanda.fetch("https://example.com", strip_mode=["js", "css"])
lightpanda.fetch("https://example.com", dump="markdown", strip_mode="clutter")

Status and multiple URLs

json=True adds the HTTP status, headers and page diagnostics:

import lightpanda

status = lightpanda.fetch("https://example.com", dump="markdown", json=True).json()
print(status["http_status"], status["content"])

results = lightpanda.fetch(["https://example.com", "https://example.org"], dump="markdown").json()
for result in results["results"]:
    print(result["url"], result["http_status"])

Any CLI flag

Underscores become dashes, True adds a bare flag, and lists repeat the flag:

import lightpanda

response = lightpanda.fetch(
    "https://example.com",
    http_header=["Authorization: Bearer token"],  # --http-header ...
    user_agent_suffix="my-crawler",               # --user-agent-suffix my-crawler
    obey_robots=True,                             # --obey-robots
    cookie_jar="cookies.json",                    # --cookie-jar cookies.json
)

Errors raise lightpanda.LightpandaError with returncode, stdout and stderr. Pass fail_on_http_error=True to also raise on HTTP 4xx/5xx.

Using Playwright or Puppeteer

import lightpanda
from playwright.sync_api import sync_playwright

proc = lightpanda.serve(host="127.0.0.1", port=9222)

with sync_playwright() as p:
    browser = p.chromium.connect_over_cdp("http://127.0.0.1:9222")
    page = browser.new_page()
    page.goto("https://example.com")
    print(page.content())
    browser.close()

proc.kill()

import lightpanda

# Write logs to files
proc = lightpanda.serve(stdout=open("lightpanda.out.log", "w"), stderr=open("lightpanda.err.log", "w"))

# CDP on / and WebDriver BiDi on /session, same port
proc = lightpanda.serve(protocol=["cdp", "webdriver"])

# CDP over a Unix-domain socket instead of TCP
proc = lightpanda.serve(unix_socket="/tmp/lightpanda.sock")

Running an MCP server

import json
import lightpanda

proc = lightpanda.mcp()

proc.stdin.write(b'{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}\n')
proc.stdin.flush()
print(json.loads(proc.stdout.readline()))  # list of available tools

proc.kill()

# Or over HTTP, clients POST JSON-RPC to /mcp
proc = lightpanda.mcp(port=8080)

Running tests

The bundled browser version is pinned in .browser-version. Download it for your platform, then run pytest:

scripts/get-browser lightpanda-x86_64-linux
uv run pytest

Available platforms:

• lightpanda-aarch64-macos - macOS arm64 (default)
• lightpanda-x86_64-macos - macOS x86_64
• lightpanda-aarch64-linux - Linux arm64
• lightpanda-x86_64-linux - Linux x86_64

License

MIT