The output is a quarter the size of the raw HTML, which is the difference between reading five pages and reading twenty in the same context. extract_url is the sharper tool: it fills a schema you define from what is actually on the page and leaves missing fields missing, rather than letting a model fill the gaps.
The official MCP server for Skim. It fetches a URL and returns the article as Markdown with a YAML frontmatter block — no navigation, ads or boilerplate — and can pull structured JSON out of a page against a schema or a preset. It also registers watches that report content diffs on later checks.
- read_url — one URL as clean Markdown with title, byline, published date, language and excerpt in frontmatter; roughly a quarter the size of the raw HTML
- read_urls — up to ten URLs in one call, returning per-URL Markdown and metadata, instead of looping read_url
- extract_url — structured JSON or tables from a page against a JSON Schema you supply, or a preset: article, product, job, review, event or table; values come only from the page
- watch_urls — register a private watch on 1 to 20 URLs and get back a watch_id to keep secret
- check_watch — poll a watch; the first check baselines each page and later ones report the content diff
A free Skim API key (it starts with sk402_) from skim402.com, set as SKIM_API_KEY; the free plan allows 1,000 reads a month and a card is required at signup. Clients run npx -y skim-mcp over stdio. An optional wallet path replaces the key entirely: set SKIM_WALLET_PRIVATE_KEY and calls pay per request in USDC on Base at $0.002 each — use a dedicated wallet, since the key sits in a plaintext config file.
One command plus a key — npx -y skim-mcp, then supply credentials