DeepScrape：开源站点转 MCP 数据源

Prasenjit Sarkar @stretchcloud原推发布于 2026.10.16

在 X 上打开原推

Firecrawl closed a $75M Series B in August, led by Smash Capital. Three months later, its newest product shipped as an MCP plugin, not an API: Fireround, a funding news feed you query straight from Claude or ChatGPT. That's the real signal, not the product itself. A scraping company with 1.5M+ developers and customers like Shopify, Apple, and Canva isn't selling an API anymore. It's turning itself into a data source an agent queries directly. That's the piece worth sitting with. Exa prices discovery by the request, 7 dollars per 1,000 searches, a dollar per 1,000 pages of content. Firecrawl prices by the credit, one per page scraped or crawled, up to 599 dollars a month for a million credits. Both models assume the agent is paying per page, every time, forever. Nobody's really asking what happens when the same hundred pages get hit daily and 90 percent of them haven't changed since yesterday. That's the gap I built DeepScrape's /api/map and change tracking around: map a site's structure once, track what actually changed, and only pull fresh data for pages that moved. Site-to-MCP turns the whole thing into a structured source an agent queries directly instead of re-scraping blind, and the output comes back as markdown, tables, or structured JSON depending on what the agent asked for, not just raw HTML dumped into a prompt. An autonomous agent can navigate toward a goal across the site instead of you writing the crawl logic by hand. Firecrawl raised $75M to build the query layer on their stack. DeepScrape is the open source bet that the pricing model underneath it, pay per page forever, is the part that should change next. https://github.com/stretchcloud/deepscrape https://x.com/firecrawl/status/2110759634164220368

这条是引用着另一条写的：x.com/firecrawl/status/2110759634164220368