MCP Server

SeaPortal runs as a Model Context Protocol server over JSON-RPC 2.0, line-delimited, on stdio:

terminal

seaportal mcp

No flags are accepted — configuration flows through tool arguments. The server implements initialize, tools/list, and tools/call.

Client config

{
  "mcpServers": {
    "seaportal": {
      "command": "seaportal",
      "args": ["mcp"]
    }
  }
}

Tools

Each tool wraps an existing library entry point and returns its JSON-marshalled result as a single text content block.

`fetch_url`

Fetch a URL and return the extracted Markdown + metadata as JSON (the Result struct). Wraps FromURLWithOptions.

Argument	Type	Description
`url`	string	required — URL to fetch
`dedupe`	boolean	Enable block dedup (default true)
`fast`	boolean	Bail early if a browser is needed
`with_links`	boolean	Emit discovered links
`with_images`	boolean	Emit discovered images
`with_tables`	boolean	Emit structured tables
`with_comments`	boolean	Emit user comments
`max_tokens`	integer	Approximate output token cap

`fetch_snapshot`

Build the accessibility-tree snapshot for a URL (HTTP fetch, then snapshot extraction). Wraps BuildSnapshotWithOptions.

Argument	Type	Description
`url`	string	required
`filter`	string	`interactive` to keep only links/buttons/inputs; empty for full tree
`max_tokens`	integer	Approximate token cap
`allow_internal`	boolean	Allow private/internal IP targets

`parse_sitemap`

Fetch and flatten a sitemap.xml (nested <sitemapindex> supported, .gz auto-decompressed). Returns a JSON array of {loc, lastmod, changefreq, priority}. Wraps FlattenSitemap.

Argument	Type	Description
`url`	string	required
`max_depth`	integer	Max sitemap-index recursion depth (default 5)
`max_urls`	integer	Stop after this many URLs (default 50000)
`allow_internal`	boolean	Allow private/internal IP targets

`parse_feed`

Fetch and parse an RSS 2.0 / Atom 1.0 / JSON Feed 1.x URL into a unified {title, link, published, summary, author, guid} JSON array. Wraps ParseFeed.

Argument	Type	Description
`url`	string	required
`max_items`	integer	Stop after this many items (default 200)
`allow_internal`	boolean	Allow private/internal IP targets

`scrape_site`

Scrape a whole site from a base URL and return a structured ScrapeResult JSON (site, pageGroups, pages, summary) designed to be handed to PinchTab for enrichment. Wraps ScrapeSite. Server-side guardrails cap max_pages (≤200), max_per_pattern (≤50), and timeout_seconds (≤180).

Argument	Type	Description
`base_url`	string	required
`max_pages`	integer	Max total pages (default 50, capped at 200)
`max_per_pattern`	integer	Max samples per URL pattern (default 8, capped at 50)
`full`	boolean	Disable sampling (fetch all discovered pages, still capped)
`include_patterns`	string	Comma-separated globs to include
`exclude_patterns`	string	Comma-separated globs to exclude
`sample_strategy`	string	`random` \| `priority` \| `balanced` (default balanced)
`with_performance`	boolean	Include per-page performance data
`respect_robots`	boolean	Respect robots.txt + crawl-delay (default true)
`timeout_seconds`	integer	Overall timeout (default 60, capped at 180)
`user_agent`	string	Override the User-Agent header
`allow_internal`	boolean	Allow private/internal IP targets

Like fetch_url, every scrape fetch (discovery, robots.txt, sitemaps, pages) runs under the secure-by-default policy; allow_internal lifts only the private-IP block.

Errors

Tool-level failures return a successful JSON-RPC result with isError: true and the error message as text; handler panics are caught and returned the same way.