fuegol.ink

The web data engine for AI. Firecrawl-compatible. Cloudflare-powered. Half the price.

Website · Console · Live API · MCP · Compatibility · Roadmap · Self-host · MIT

---

fuegol.ink is an open-source, Cloudflare-native web-data platform that speaks the Firecrawl v2 API. Point an existing Firecrawl client at api.fuegol.ink, swap in a fuegol key, and keep your code. The managed edition targets ≈50% of Firecrawl's published plans; the engine is MIT and self-hostable on your own Cloudflare account.

Compatibility means: change your base URL + use a new fuegol key (never your fc- key). Your request/response business logic stays the same.

30-second quickstart

The public demo is live and keyless (rate-limited). Scrape any page to clean Markdown:

curl -X POST https://fuegol-api.manhattan.workers.dev/v2/scrape \
  -H 'content-type: application/json' \
  -d '{"url":"https://example.com","formats":["markdown","links"]}'

Real response (trimmed):

{
  "success": true,
  "data": {
    "url": "https://example.com/",
    "title": "Example Domain",
    "markdown": "This domain is for use in documentation examples without needing permission. This is not a service; avoid relying on it for testing and monitoring purposes.",
    "links": [],
    "metadata": {
      "statusCode": 200,
      "title": "Example Domain",
      "language": "en",
      "sourceURL": "https://example.com/",
      "contentType": "text/html; charset=utf-8"
    }
  }
}

_{Verbatim from the live endpoint on 2026-10-07 (example.com was redesigned — it no longer links out).}

Map a site's URLs:

curl -X POST https://fuegol-api.manhattan.workers.dev/v2/map \
  -H 'content-type: application/json' -d '{"url":"https://example.com","limit":50}'

How it works

A request enters a cheapest-adequate-tier planner. Most pages never touch a browser — they are served by a free static path (SSRF-guarded fetch → readability extraction → HTML→Markdown). The browser tier is reserved for pages that truly need JS, actions, screenshots, or an anti-bot proxy.

flowchart LR
  C[Client / SDK / MCP] -->|Firecrawl v2| W[Workers + Hono API]
  W --> P{Execution planner}
  P -->|static, free| H[fetch → readability → HTML→Markdown]
  P -->|JS / screenshot / stealth| B[Browser Rendering]
  P -->|breadth| CR[/crawl async/]
  P -->|structured| AI[Workers AI + AI Gateway]
  H --> N[Normalize → Firecrawl Document]
  B --> N
  CR --> N
  AI --> N
  N --> R[(R2 artifacts)]
  W --> D[(D1 keys + credit ledger)]
  W --> Q[[Queues → Durable Object crawl]]
  N --> C

Loading

Edge primitives: Workers + Hono · Browser Rendering · Queues · Workflows · Durable Objects · D1 · R2 · Workers AI · AI Gateway · AI Search. No portability layer — the CF integration is the cost and latency advantage. See docs/architecture-decisions.md.

What works today

Every row below is live on the production API and covered by a reproducible prod E2E in e2e/.

Capability	Status
POST /v2/scrape → markdown / html / rawHtml / links / summary / metadata	✅ live
POST /v2/scrape → {type:"json"} AI extraction (Workers AI)	✅ live
POST /v2/scrape → JS-rendered (Browser Rendering) + screenshot → R2	✅ live
POST /v2/scrape → actions (click/write/press/scroll/wait/screenshot/scrape/executeJavascript/pdf)	✅ live
POST /v2/scrape → changeTracking (git-diff + AI json semantic diff)	✅ live
POST /v2/scrape → maxAge result cache (R2, content-addressed) + storeInCache	✅ live
POST /v2/extract async structured extraction (Workers AI, Durable Object)	✅ live
POST /v2/map → sitemap + link discovery	✅ live
POST /v2/crawl + POST /v2/batch/scrape async (Durable Object) + status/cancel/errors + signed webhooks	✅ live
POST /v2/search → multi-source (web / news / images) + categories + tbs time-filter + optional result-scraping (Exa/Tavily)	✅ live
POST /v2/parse — PDF (unpdf) + HTML/text → markdown	✅ live
POST /v2/agent — autonomous research (search → scrape → synthesize + sources)	✅ live
POST /v2/monitor (+ run/checks) — recurring change detection on Cron + signed monitor.changed webhooks	✅ live
POST /v2/browser + /browser/:id/act — persistent interactive browser sessions (state persists across requests)	✅ live
POST /v2/ai-search/index + /query — AI Search: semantic search + RAG over your indexed content (Vectorize + Workers AI, per-key)	✅ live
API keys (POST /v2/keys) + D1 credit ledger, enforced spend ceilings (402) + user spend limits	✅ live
SSRF guard (v4/v6/metadata/redirect/DoH), robots.txt, honest error envelopes, /v1/* adapters, /concurrency-check	✅ live
Remote MCP — scrape/map/crawl/search + developer/gov/research search + agent + monitor_* + research inspect/related/read	✅ live
Dev console /app/ — keys, live credits, playground (scrape/map/search/crawl) + interactive-session playground	✅ live
Stripe billing · custom domains · one-click Deploy button	⛔ roadmap (needs a Stripe test key / the DNS zone)
find_tools (Alexandria data-provider catalogue)	⛔ proprietary upstream

Full surface map: docs/product-surface-inventory.md. Every unbuilt endpoint returns an honest 501 with a pointer — never a fake success object. Real wall-clock numbers (static scrape P50 ≈ 100 ms, browser tier ≈ 1.5 s) live in docs/benchmarks.md, reproducible via node e2e/benchmark/run.mjs.

Firecrawl compatibility

Pinned to the real upstream source (there is no committed v2 OpenAPI — the contract lives in Firecrawl's Zod, which we mirror in packages/contracts):

Surface	Pinned revision
REST API v2	firecrawl/firecrawl@f9f2e3d
MCP server	firecrawl-mcp-server@ec0f9de (firecrawl-mcp@3.28.2) — 30 tools registered / 28 listed

The MCP server's README says "27 tools"; the source at HEAD registers 30 (28 listable + 2 hidden deprecated shims). We pin to the source count. Details: docs/firecrawl-compatibility.md.

Pricing (managed edition — Stripe not yet live)

Plans are exactly half of Firecrawl's month-to-month price, like-for-like credits (re-verified 2026-10-09). The "50% less" claim is measured against Firecrawl's month-to-month price; their annual-billed equivalents ($16/$83/$333/$599) are shown too, and fuego is cheaper even against those. Live price book: GET /v2/pricing. Preflight a job cost: POST /v2/quote.

Tier	Monthly credits	Firecrawl (month-to-month)	fuego (½)
Free	1,000	$0	$0
Hobby	5,000	$19	$9.50
Standard	100,000	$99	$49.50
Growth	500,000	$399	$199.50
Scale	1,000,000	$749	$374.50

One-time credit packs (no subscription): $5→2,000 · $10→4,500 · $25→12,000 · $50→26,000 · $100→55,000 · $250→145,000 · $500→300,000. Core ops bill at Firecrawl parity; Fuego-exclusive premium ops (browser/AI/agent) are priced to a proven ≥50% margin floor against the worst-case lot — never a half-price loss leader. Cost model + margin proof: docs/unit-economics.md · status gates: docs/RELEASE_CHECKLIST.md. Billing (subscriptions + packs) is built behind the ledger and resumes the moment a Stripe sk_test_ key is provisioned.

Examples

TypeScript

const res = await fetch('https://fuegol-api.manhattan.workers.dev/v2/scrape', {
  method: 'POST',
  headers: { 'content-type': 'application/json' /* , authorization: `Bearer ${fuegolKey}` */ },
  body: JSON.stringify({ url: 'https://example.com', formats: ['markdown', 'links'] }),
});
const { data } = await res.json();
console.log(data.markdown);

Python

import requests
r = requests.post("https://fuegol-api.manhattan.workers.dev/v2/scrape",
                  json={"url": "https://example.com", "formats": ["markdown"]})
print(r.json()["data"]["markdown"])

CLI (@fuegol/cli — zero-dependency)

fuegol scrape https://example.com
fuegol search "durable objects" --category developer
FUEGOL_API_KEY=fgl_live_… fuegol crawl https://docs.site --limit 20 --wait

MCP — live at https://fuegol-mcp.manhattan.workers.dev. Real working tools: firecrawl_scrape, firecrawl_map, firecrawl_crawl, firecrawl_check_crawl_status, firecrawl_search (web/news/images), firecrawl_developer_search, firecrawl_gov_search, firecrawl_research_search_papers, firecrawl_research_inspect_paper / _related_papers / _read_paper (Semantic Scholar), firecrawl_agent, and firecrawl_monitor_create / _list / _run / _checks (the MCP forwards your Bearer key so authed tools hit your ledger). Only firecrawl_find_tools (proprietary) is a stub. Stateless Streamable-HTTP JSON-RPC.

{ "mcpServers": { "fuegol": { "url": "https://fuegol-mcp.manhattan.workers.dev/v2/mcp" } } }

# verify it live:
curl -X POST https://fuegol-mcp.manhattan.workers.dev/v2/mcp -H 'content-type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"firecrawl_scrape","arguments":{"url":"https://example.com"}}}'

Interactive browser sessions — drive a persistent headless browser; the tab's state (URL, cookies, form input) survives across separate requests:

# start a session, then click "next" + screenshot — the tab persists between calls
ID=$(curl -s -X POST https://fuegol-api.manhattan.workers.dev/v2/browser \
  -H 'content-type: application/json' -d '{"url":"https://quotes.toscrape.com/"}' | jq -r .id)
curl -X POST https://fuegol-api.manhattan.workers.dev/v2/browser/$ID/act \
  -H 'content-type: application/json' \
  -d '{"actions":[{"type":"click","selector":"li.next a"},{"type":"screenshot"}]}'

sequenceDiagram
  participant C as Client
  participant W as Workers API
  participant DO as BrowserSession DO
  participant BR as Browser Rendering
  C->>W: POST /v2/browser {url}
  W->>DO: create()
  DO->>BR: launch(keep_alive) → navigate
  DO-->>C: { id, url, title }
  Note over C,BR: later request — tab still alive
  C->>W: POST /v2/browser/:id/act {actions}
  W->>DO: act(actions)
  DO->>BR: reconnect(sessionId) → reattach live tab
  BR-->>DO: click / type / screenshot
  DO-->>C: { url, actCount, screenshot→R2 }

Loading

Or drive it visually in the console playground.

Migrating from Firecrawl

- const base = 'https://api.firecrawl.dev';
- const key  = 'fc-...';
+ const base = 'https://api.fuegol.ink';   // or your self-hosted Worker
+ const key  = 'fgl-...';                  // a NEW fuegol key — fc- keys are not accepted

Or use the official Firecrawl SDK unchanged — just point it at fuegol:

import { Firecrawl } from '@mendable/firecrawl-js';
const app = new Firecrawl({ apiKey: 'fgl_live_...', apiUrl: 'https://api.fuegol.ink' });
const doc = await app.scrape('https://example.com', { formats: ['markdown'] });

Verified against both official SDKs: @mendable/firecrawl-js@4.45.0 8/8 and firecrawl-py@4.49.3 4/4 run unchanged against fuegol.ink — scrape · map · search · crawl (+ batch/extract/status in TS). Reproduce: node e2e/sdk-compat/test.mjs · python e2e/sdk-compat/test.py.

Your scrape/map/crawl request bodies and response handling stay the same (within the supported surface above). A fuegol key is required — an existing Firecrawl key is never automatically valid.

Self-hosting

MIT-licensed; runs on your Cloudflare account with no Stripe and no dependency on fuegol.ink infrastructure. The free static tier needs zero extra provisioning. See deploy/README.md.

git clone https://github.com/heymegabyte/fuegol.ink.git && cd fuegol.ink
pnpm install && cd apps/api && pnpm exec wrangler deploy

Deploy-to-Cloudflare button: Cloudflare's one-click button can't resolve a pnpm monorepo's workspace packages at build time (documented limitation). The dependency-isolated standalone Worker for one-click deploy is a tracked increment — we won't show a "working" badge until it's tested from a clean account. The manual command above is verified.

Security & privacy

• SSRF-hardened fetch: blocks private/reserved IPv4+IPv6, cloud-metadata endpoints, non-http(s) schemes, and re-validates every redirect hop; optional DNS-over-HTTPS rebinding check.
• Respects robots.txt and Cloudflare Content Signals. No CAPTCHA bypass, credential theft, private- network scanning, or paywall circumvention.
• Signed webhooks (HMAC-SHA256, x-fuegol-signature) fire on crawl/batch completion and on monitor change detection (monitor.changed), with retries. Webhook targets are SSRF-guarded.
• Zero-data-retention mode, PII redaction, and per-tenant isolation are on the roadmap.
• Observability: server-side Sentry (@sentry/cloudflare) on the API + MCP Workers — errors + traces, with every Durable Object instrumented — plus Workers Tracing. No browser SDK.

Repository layout

packages/contracts  Zod SSOT — Firecrawl v2 request/response schemas (the spec mirror)
packages/engine     Cloudflare-native engine — SSRF, fetch, extraction, HTML→MD, planner
apps/api            Hono Worker — the Firecrawl-compatible REST API (live)
docs/               idea-ledger · compatibility · surface-inventory · unit-economics · ADRs · roadmap · convergence-log
deploy/             self-hosting guide + standalone-template plan

Contributing

Issues and PRs welcome. Run pnpm install, then pnpm -r typecheck and pnpm --filter @fuegol/engine test. Add a regression test with every bug fix; keep the Zod contracts authoritative.

License

MIT © 2026 Megabyte Labs. Not affiliated with Firecrawl; "Firecrawl" is used only to describe API compatibility. A clean-room implementation pinned to public upstream revisions.

---

Point your client at fuegol.ink. Keep your code. Pay half.