$sudofetch

§ 1  Web retrieval for agents and pipelines

The pages that usually fail.

One request returns clean Markdown, HTML or text from a real browser. When a site blocks, sudofetch climbs to a residential connection, then a hardened browser, and charges only for the level that got through.

Get an API key Read the docs300 trial credits · failed requests are free
Request slip · form 07Cleared · L2
POST https://api.sudofetch.com/v1/scrape
{ "url": "https://shop.example/p/linen-shirt",
  "format": "markdown" }
status
succeeded
mode_used
residential
attempts
2
credits.charged
15

# Level 1 was refused (403). Level 2 returned the page. You paid for the level that worked.

§ 2  Clearance levels

Escalate only when refused.

Every job starts at the cheapest level. A recognized block, challenge or bare 401/403 moves it one level up, once, within the time and credit ceiling you set. Pages that simply don't exist, and real logins, stop immediately.

  1. LEVEL1

    Direct

    A clean hosted browser on a datacenter connection. Most pages end here.

  2. LEVEL2

    Residential

    The same browser on a US residential connection, after a recognized block.

  3. LEVEL3

    Premium

    A hardened browser with challenge solving, when residential is refused too.

§ 3  Public benchmark

73.8%

369/500 · rank 7 of 17

On the Web Data Frontier benchmark's 100 hard targets, graded the same way as the 16 other scraping providers in its published results.

100 targets × 5 attempts, 90 s deadline, HTML output, auto mode, 60-credit ceiling, concurrency 4. Field results from the benchmark's official 2026-09-15 run. Method and targets.

Compatibility · passes of 500field 2026-09-15 · ours 2026-09-23
  1. 01String97.0%
  2. 02Scrapfly86.2%
  3. 03ScraperAPI84.0%
  4. 04Firecrawl80.2%
  5. 05Apify77.4%
  6. 06Bright Data74.6%
  7. 07sudofetch73.8%
  8. 08ScrapingBee73.0%
  9. 09Context.dev72.0%
  10. 10Oxylabs69.0%
  11. 11Nimble68.6%
  12. 12Zyte68.0%
  13. 13Decodo50.6%
  14. 14Scrapingdog45.6%
  15. 15Browserbase41.4%
  16. 16ZenRows41.2%
  17. 17ScrapingAnt36.4%

§ 4  Built for agents

01

Clean output

Markdown, HTML or text from the page's main content, with title and metadata. No navigation soup.

02

Durable jobs

Idempotent submissions, a synchronous wait of up to 60 s, and polling for anything longer.

03

Metered, not guessed

A base price per level plus transfer beyond an allowance. Your credit ceiling caps both.

04

MCP included

Drop the MCP server into your agent and call scrape. Content arrives marked as untrusted data.

Pay per page that worked. Failed requests cost nothing.

See pricing