This is the English part of the site. Sign-up, dashboard, checkout and legal texts are in German – the API, the SDKs and these pages are all you need to try it.

Developers

Developer quickstart

Base URL https://api.ragfeed.de. Authenticate with Authorization: Bearer sk-… – keys always start with sk-. The full interactive reference is at /docs (field descriptions partly in German).

Quickstart

  1. Get a key. Sign up with your e-mail address at /signup (German form: enter your address, click the confirmation link). Your key is shown once and never sent by e-mail.
  2. First request. 
    curl -X POST https://api.ragfeed.de/v2/scrape \
  -H "Authorization: Bearer sk-..." \
  -H "Content-Type: application/json" \
  -d '{"url": "https://example.com", "formats": ["llmMarkdown"]}'
    Response
    {
  "success": true,
  "data": {
    "llmMarkdown": "---\ntitle: Example Domain\nurl: \"https://example.com/\"\nlanguage: en\ntokens: 279\n---\n\nThis domain is for use in documentation examples …",
    "metadata": {
      "title": "Example Domain",
      "language": "en",
      "url": "https://example.com/",
      "statusCode": 200,
      "contentType": "text/html",
      "creditsUsed": 1
    }
  }
}
  3. Keep going. Switch existing Firecrawl code (below), get chunks for your vector database (formats) or connect a tool without code (integrations).

Existing Firecrawl code: change one line

The API is compatible with the open-source SDKs firecrawl-py and @mendable/firecrawl-js. Set api_url / apiUrl to https://api.ragfeed.de and use your Ragfeed key; the rest of your code stays.

Python:

# pip install firecrawl-py
from firecrawl import Firecrawl

app = Firecrawl(api_key="sk-...", api_url="https://api.ragfeed.de")   # the only change
doc = app.scrape("https://example.com", formats=["markdown"])

Node.js:

// npm i @mendable/firecrawl-js
import Firecrawl from "@mendable/firecrawl-js";

const app = new Firecrawl({ apiKey: "sk-...", apiUrl: "https://api.ragfeed.de" });   // the only change
const doc = await app.scrape("https://example.com", { formats: ["markdown"] });

The official SDKs don't know llmMarkdown and chunks and may reject them before sending – use plain HTTP for those two formats.

Formats: llmMarkdown and chunks

  • llmMarkdown – Markdown of the main content with a header (title, URL, language, fetch time, token count); link targets as a numbered list at the end, images as alt text only.
  • chunks – the page split along its headings into ready records for a vector database. Each chunk has a stable UUID (same page + position = same id, so re-fetching overwrites instead of duplicating), the heading path, text and embedText, token count, hash and flat metadata that fits Pinecone, Qdrant, Weaviate, Milvus, Chroma and pgvector. You create the vector with your own embedding model.
{
  "url": "https://example.com/guide",
  "formats": ["llmMarkdown", {"type": "chunks", "model": "openai-3", "profile": "balanced"}]
}
OptionEffect
modelYour embedding model: openai-3, mistral-embed, multilingual-e5, bge-m3, jina-v3, cohere-v3, gemini. Every chunk then fits its input window.
profilebalanced (500 tokens, default), precise (250), broad (1,000)
maxTokens, overlapUpper limit per chunk (100 to 4,000) and tokens repeated from the previous chunk
contextWhat embedText puts before the text: title, headings, description. Default: nothing.
metadataUp to 20 own fields per chunk, e.g. category or tenant for filtering
idNamespaceOwn namespace for ids, so the same page gets different ids in different projects
shapelangchain → Document(**chunk), llamaindex → TextNode(**chunk)

Both formats cost no extra credits.

Endpoints

Firecrawl API v2; the v1 paths work as well.

EndpointWhat for
POST /v2/scrapeOne page as Markdown, llmMarkdown, chunks, HTML, links, screenshot or JSON. JavaScript is rendered, PDFs come back as text.
POST /v2/crawlWhole sites incl. sitemap, with path filters, depth limit and per-page webhooks
POST /v2/batch/scrapeMany URLs in one job
POST /v2/mapAll URLs of a domain, optionally ranked by a search term
POST /v2/searchWeb search, optionally with the results fetched
POST /v2/extractStructured data following your JSON schema, also across many pages
POST /v2/parseDocument upload: PDF, Office, OpenDocument, EPUB, CSV
POST /v2/monitorMonitoring Beta, from plan Fluss: check pages on a schedule and report changes, incl. which chunks changed

Integrations

ToolHow
Claude / any MCP clienthttps://api.ragfeed.de/mcp, OAuth 2.1 with dynamic client registration, or your API key as Bearer. Claude Code: claude mcp add --transport http ragfeed https://api.ragfeed.de/mcp
n8nOfficial Firecrawl node, credential Base URL https://api.ragfeed.de/v2 + your sk- key – or an HTTP Request node on /v2/scrape
LangChainFireCrawlLoader(api_url=…); chunks with "shape": "langchain" → Document(**chunk)
LlamaIndexChunks with "shape": "llamaindex" → TextNode(**chunk)
DifyKnowledge base, data source Firecrawl, base URL https://api.ragfeed.de
Make“HTTP – Make a request” with header Authorization: Bearer sk-…

Limits and credits

PlanCredits / monthConcurrent requestsRequests / minPrice / month
Quelle (free)1,0002 10€0
Bach5,0005 60€9
Fluss100,00015 300€49
Strom500,00030 600€199

A page costs 1 credit (Markdown, llmMarkdown, chunks, HTML, links, screenshot), every AI format 5 credits extra. Failed pages cost nothing. GET /v2/team/credit-usage returns your remaining credits. Over the per-minute limit you get 429 RATE_LIMIT_EXCEEDED with the wait time; out of credits, 402.

What's not supported

Formats branding, audio, video; persistent browser sessions. Sites that block data-centre traffic stay blocked – we don't bypass logins, paywalls or CAPTCHAs. No SLA.

More (German)

The full German documentation covers crawl options, webhooks, document parsing, zero data retention and the monitoring beta with chunkDiff: /entwickler.