SeaPortal

HTTP-first web content extraction for AI agents. Turn static and server-rendered pages into clean Markdown or a JSON accessibility snapshot, parse sitemaps and feeds, and run it as a CLI, a Go library, or an MCP server — secure by default, with an explicit signal when a page actually needs a browser.

Install

npm install -g seaportal
# or
npx seaportal <url>

Usage

# Basic extraction
seaportal https://pinchtab.com

# JSON output
seaportal --json https://pinchtab.com

# Fast mode (bail early if browser needed)
seaportal --fast https://pinchtab.com

# Disable deduplication
seaportal --no-dedupe https://pinchtab.com

# Combine options
seaportal --json --fast https://pinchtab.com

Output

SeaPortal outputs Markdown with YAML frontmatter containing metadata:

---
title: "Page Title"
url: https://pinchtab.com
confidence: 85
isSpa: false
needsBrowser: false
---

# Page Title

Content extracted as clean Markdown...

Features

• Fast — Pure HTTP, no browser required (<2s typical on static / SSR pages)
• Clean Markdown — Readability extraction + block and simhash dedupe
• Accessibility snapshots — JSON semantic tree with role, ref, and selector per node
• Browser-needed signal — Flags SPA / blocked pages so you can route them elsewhere
• Sitemaps & feeds — Flatten sitemap.xml and parse RSS, Atom, and JSON Feed
• MCP server — seaportal mcp exposes fetch_url, fetch_snapshot, parse_sitemap, and parse_feed over stdio
• Safe by default — SSRF / private-IP blocking, http(s)-only, redirect and body caps on the CLI and MCP server

Environment Variables

• SEAPORTAL_BINARY_PATH — Custom binary path (for Docker, dev builds)

License

MIT