Provider benchmark · web.scrape.markdown
Scrape a web page into markdown
7 providers ranked for web.scrape.markdown on 25,691 real looot calls in the last 30 days. Steel API has the highest success rate, 95.2%. Jina AI costs least per success, $0.00021.
Updated
- Providers
- 10
- Ranked providers
- 7
- 20+ calls in 30 days
- Calls in 30 days
- 25,691
- Top success rate
- 95.2%
- Steel API
- Fastest median
- 482 ms
- Keenable
- Cheapest per success
- $0.00021
- Jina AI
Leaderboard
Who does web.scrape.markdown best
Ranked by success rate, then cost per success, then speed. A provider needs 20 calls in the last 30 days to be ranked.
| Rank | Provider | Success rate | Median time | Price | Cost per success | Calls (30 days) |
|---|---|---|---|---|---|---|
| 1 | Steel API | 95.2% | 5.9 s | $0.005 per call | $0.00525 | 84 |
| 2 | Jina AI | 93.6% | 4.9 s | $0.0002 per call | $0.00021 | 24,416 |
| 3 | Context.dev | 90.8% | 8.4 s | $0.0025 per call | $0.00275 | 756 |
| 4 | You.com | 87.8% | 886 ms | $0.001 per result | Depends on results | 314 |
| 5 | Olostep | 82.2% | 6.9 s | $0.0018 per call | $0.00219 | 45 |
| 6 | Keenable | 76.0% | 482 ms | $0.004 per call | $0.00526 | 25 |
| 7 | ScrapeGraphAI | 4.5% | 1.0 s | $0.012 per call | $0.264 | 22 |
| Not enough calls to rank | Serper | 86.6% | 705 ms | $0.002 per call | Not ranked | 15 |
| Not enough calls to rank | Limadata | Not published | Not published | $0.004 per result | Not ranked | Under 10 |
| Not enough calls to rank | Scrapfly | Not published | Not published | $0 per call | Not ranked | Under 10 |
Run it
Run web.scrape.markdown from your agent
Run the job itself (job:web.scrape.markdown) with fallback on: looot picks a provider that can answer now and tries the next on a miss. You see the price before each call and pay from a prepaid balance.
No looot yet? Connect the MCP server below, or start a workspace.
Connect in 30 seconds
Pick the client you use and add looot once. Each client takes one step, then your agent does the rest.
01Claude Code
Run this in your terminal, then sign in from the browser when it asks.
Terminal commandclaude mcp add looot https://api.looot.ai/mcp --transport http
02Cursor
Create .cursor/mcp.json with this server, then sign in from the browser.
mcp.json{"mcpServers":{"looot":{"url":"https://api.looot.ai/mcp"}}}
03ChatGPT
Where your plan lets you add a connector, add this URL and sign in.
Connector URLhttps://api.looot.ai/mcp
Then tell your agent, in any client:
set up https://looot.ai/skill.mdEndpoints
Every endpoint for web.scrape.markdown
Each endpoint in the public catalog that does this job, with its own numbers. Numbers appear from 10 calls in the last 30 days.
| Endpoint | Provider | Success rate | Median time | Price | Calls (30 days) |
|---|---|---|---|---|---|
| Jina AI / Read a web page as markdown | Jina AI | 93.6% | 4.9 s | $0.0002 per call | 24,416 |
| Context.dev / Scrape a web page to markdown | Context.dev | 90.8% | 8.4 s | $0.0025 per call | 756 |
| You.com / Page contents | You.com | 87.8% | 886 ms | $0.001 per result | 314 |
| Steel API / Scrape a web page to markdown or HTML | Steel API | 95.2% | 5.9 s | $0.005 per call | 84 |
| Olostep / Scrape page | Olostep | 82.2% | 6.9 s | $0.0018 per call | 45 |
| Keenable / Fetch page as markdown | Keenable | 76.0% | 482 ms | $0.004 per call | 25 |
| ScrapeGraphAI / Scrape page (stealth) | ScrapeGraphAI | 4.5% | 1.0 s | $0.012 per call | 22 |
| Serper / Scrape web page | Serper | 86.6% | 705 ms | $0.002 per call | 15 |
| ScrapeGraphAI / Scrape page | ScrapeGraphAI | 78.5% | 1.1 s | $0.002 per call | 14 |
| Limadata / Extract clean markdown from web pages | Limadata | Not published | Not published | $0.004 per result | Under 10 |
| Scrapfly API / Scrape a web page | Scrapfly | Not published | Not published | $0 per call | Under 10 |
How we measure
- Window: every call made through looot in the last 30 days, read 2026-11-10 23:26 UTC.
- Success rate: the share of calls the provider answered without an error. An empty answer (for example no email found) counts as an answer. It does not check that the data is right. Calls that failed on looot's side are left out.
- Median time: half of the calls answered faster than this, as looot measured them. A 95th percentile is not published yet.
- Minimum calls: an endpoint shows numbers from 10 calls, and a provider is ranked from 20 calls. Rows below that say "Not enough calls to rank".
- One row per provider: the provider's endpoint with the most calls for the job. Every endpoint is listed in the endpoints table.
- Price: what looot charges per call or per result, in US dollars, from a prepaid balance. No subscription and no credits.
- Cost per success: the price per call divided by the success rate. A per-result price depends on how many results a call returns, so it has no cost per success.
- Source: GET https://api.looot.ai/v1/public-catalog/<endpoint>?stats=1 for each endpoint, and the public catalog for prices and jobs.
FAQ
Questions
Run web.scrape.markdown on the provider that answers.
One key, one prepaid balance, the price shown before each call.