Web data API for AI agents: search, scrape, crawl and interact, returning LLM-ready Markdown. Open source core with a free plan and an MCP server.
Best for developers who give an agent web access and start from URLs rather than questions. The free plan allows 2 concurrent requests, so it suits prototypes. Watch the credit cost of structured JSON output before you scale, and note that privacy features are Enterprise only.
Search, scrape, crawl and page interaction behind one credit balance and one MCP server.
Default placement in Hermes Agent and a hosted MCP server create habit and distribution, while scraping and search are crowded and open source lets rivals copy the core.
Main trade-off
Credit costs rise sharply once structured JSON extraction is on, and privacy controls sit behind Enterprise.
| Plan | Monthly price | What it includes |
|---|---|---|
| Free | Free | 1,000 credits a month, no card, 2 concurrent requests, 10 scrape or search calls a minute, community support |
| Hobby | $16/mo | 5,000 credits a month, 5 concurrent requests, 100 calls a minute, basic support, $16 billed yearly or $19 billed monthly |
| Standard | $83/mo | 100,000 credits a month, 25 concurrent requests, 500 calls a minute, standard support, billed yearly |
| Growth | $333/mo | 500,000 credits a month, 50 concurrent requests, 5,000 calls a minute, priority support, shared Slack channel, billed yearly |
| Scale | $599/mo | 1,000,000 credits a month, 100 concurrent requests, 10,000 calls a minute, one month credit rollover, invoiced billing, billed yearly |
| Enterprise | Custom | Custom credits and concurrency, SSO and SCIM, zero data retention, dedicated support with SLA, bulk discounts |
Free plan with 1,000 credits a month. Paid plans billed yearly: Hobby $16/mo (5,000 credits), Standard $83/mo (100,000), Growth $333/mo (500,000), Scale $599/mo (1,000,000). Hobby is $19/mo billed monthly. Enterprise is custom with zero data retention and SSO. Pay as you go adds credits in $5 steps on paid plans. 1 credit = 1 page on a basic scrape, crawl or map; search is 2 credits per 10 results; Interact is 2 credits per browser minute. Prices in USD, effective 2026-09-04.
Firecrawl is an open source web data API that turns search, scrape, crawl and browser interaction into LLM-ready Markdown or JSON. Its core repository has 188,100 GitHub stars under AGPL-3.0, and the company reports more than 1.5 million users. It is the default web search provider in Hermes Agent.
Search with page content
One search call returns web, image and news results with full page content, billed at 2 credits per ten results.
Scrape to Markdown or JSON
Converts any URL to Markdown, HTML, screenshots or structured JSON, with proxies, anti-bot handling and JavaScript rendering handled for you.
Crawl and Map
Walks a whole site or lists its URLs at 1 credit per page, with webhooks for long jobs.
Interact with live pages
Continues from a scraped page to click, fill forms and extract dynamic content, billed at 2 credits per browser minute.
Hosted MCP server and CLI
Adds to Claude Code, Codex, Cursor or OpenCode with one command against mcp.firecrawl.dev, with browser sign-in.
Alexandria data sources
Combines official data providers, custom connectors and Firecrawl indexes with the live web behind one interface for agents.
| Tool | Price from | Free plan | Runs on | Rating |
|---|---|---|---|---|
| $16/mo | Yes | Web | -- | |
| $49/mo | Yes | Web | 4.0 | |
| $17/mo | Yes | Web · iOS · Android · Mac · Windows | 4.4 | |
| -- | Yes | Web | 3.9 |
Open source core under AGPL-3.0; SDKs and some UI components under MIT. The cloud service adds features not in the repository.
Quoted per month when billed yearly: Hobby $16 (5,000 credits), Standard $83 (100,000), Growth $333 (500,000) and Scale $599 (1,000,000). Hobby costs $19 when billed monthly, and Enterprise is custom. Paid plans can top up with pay as you go. Structured output formats add extra credits per page.
The Free plan includes 1,000 credits that refresh every month, with no credit card. That covers about 500 searches or 1,000 scraped pages, at 2 concurrent requests and 10 scrape or search calls a minute. SOC 2 Type II coverage, PII redaction and MFA apply to every plan, but zero data retention does not.
Pick Exa when you want its own neural index and ranked retrieval rather than page scraping. Pick Perplexity when you need cited, synthesized answers instead of raw content. You.com is a third search API for agents, and the Linkup skill covers search with sourced results. None of them crawl or interact with a live page the way Firecrawl does.
Firecrawl is the better fit when the job starts from a URL: scraping, crawling a site, filling a form, then returning Markdown. Exa is stronger when the job starts from a question and you want ranked results from an index it built. Both appear in the keyless ring Hermes Agent uses when no web key is set, so many builds use one for search and the other for extraction.
Send a POST to the v2 scrape endpoint and you get a result without an account or key, with lower limits. For real use, create a key on the free plan and install an SDK for Python or Node. For agents, one command adds the hosted MCP server to Claude Code or Codex, then you sign in through the browser. Set FIRECRAWL_API_KEY if your agent framework expects it.
Firecrawl is a web data API built for AI agents. You send it a query or a URL and get back clean Markdown or structured JSON instead of raw HTML. The core repository is open source under the AGPL license, and the company says more than 1.5 million users build with it. The team previously made the documentation chat product Mendable, and announced a $75M Series B led by Smash Capital on 2026-09-22.
The product is a set of endpoints that share one credit balance. Search returns results with full page content in one call. Scrape converts any URL to Markdown, HTML or JSON. Crawl and Map walk a whole site, Parse handles documents, Interact lets you click and fill forms on a page that is already scraped, and Agent runs research tasks with dynamic pricing. Proxies, anti-bot handling and JavaScript rendering are handled on the vendor side, so your code sees one response shape.
It is most useful as the web layer under an agent. Hermes Agent ships it as the default provider for web search and page extraction, and OpenClaw builds can use it in the same slot. A hosted MCP server at mcp.firecrawl.dev connects it to Claude Code and Cursor in one command, and the SDKs plug into LangChain, LlamaIndex and n8n pipelines. For search-only needs, compare Exa, Perplexity and You.com, or the Linkup skill.
Pricing is credit based. A basic scrape, crawl or map costs 1 credit per page, search costs 2 credits per ten results, and Interact costs 2 credits per browser minute. The Free plan needs no card, and paid plans run from Hobby to Scale, with Enterprise adding zero data retention and SSO. Hermes Agent's own docs quote a 500-credit free tier, while Firecrawl's pricing page lists a larger allowance, so this page follows the vendor's own figure. You can also self-host the open source core, but the cloud adds features the repository does not include.
Product facts on this page (pricing, plans, integrations, compliance) are read from Firecrawl's own pages and documentation. The verdict is Hokai's reading of them, not the vendor's claim.