jina-reader

Jina Reader (zero-RAM page fetch)

Fetch one URL as clean markdown with a single HTTP GET — no MCP server, no subprocess, no RAM footprint. This is the leanest fetch in the stack; reach for it before waking a heavier tool.

When to use

• A simple / mostly-static page: article, blog post, docs page, README, changelog, release notes.
• You need the readable text/markdown, not screenshots, brand extraction, crawling, or a schema.
• You want to avoid spawning a resident MCP process (RAM-conscious sessions).

Prefer a heavier tool when: the page is JS-heavy / renders client-side, you need to crawl or map a site, you need structured/schema extraction or screenshots (use firecrawl), or you want to batch many URLs in one call (use the exa MCP web_fetch_exa).

How (HTTP GET — pick your client)

Prefix the target URL with https://r.jina.ai/:

# Bash / curl
curl -sS "https://r.jina.ai/https://example.com/some/article"

You can also use the WebFetch tool against https://r.jina.ai/<TARGET_URL> — same idea, no shell.

Optional headers

• Authorization: Bearer $JINA_API_KEY — higher rate limits + the free 10M-token allowance. Read the key from the environment; never hardcode or print it. Skip the header entirely for anonymous use.
• X-Return-Format: markdown (default) — also text, html.
• X-Target-Selector: <css> / X-Remove-Selector: <css> — keep/drop parts of the DOM.
• X-With-Links-Summary: true — append a consolidated list of the page's links.

Authenticated (key from env — never inline the literal)

Fallback chain

Guardrails