Spicrawl TypeScript SDK

The Spicrawl TypeScript SDK (@spicrawl/sdk) is the official Node.js client for the Spicrawl web scraping API: it turns any URL into clean Markdown, HTML, JSON, a screenshot or a PDF, with JavaScript rendering, structured data extraction and batch jobs, for developers building AI agents, RAG pipelines and LLM apps.

Docs · Get an API key · CLI · MCP server · Agent plugins · Changelog

Install

npm install @spicrawl/sdk
pnpm add @spicrawl/sdk
bun add @spicrawl/sdk

Create an API key at app.spicrawl.com and export it. Keys start with spicrawl_live_ (spends credits) or spicrawl_test_ (never spends live credits).

export SPICRAWL_API_KEY=spicrawl_live_...

Quickstart: scrape a URL to Markdown

// quickstart.mjs: run with `node quickstart.mjs`
import { Spicrawl } from "@spicrawl/sdk";

const spicrawl = new Spicrawl(); // reads SPICRAWL_API_KEY

const page = await spicrawl.scrapeRaw({
  url: "https://example.com",
  response_format: "markdown",
  main_content_only: true, // drop nav, footer and sidebars
});

if (page.format === "markdown") console.log(page.body);
console.log(page.targetStatus, page._meta.engine, page._meta.creditsCharged);

It prints the page as Markdown, then the target site's status, the engine that ran and the credits charged (a plain fetch scrape costs 1 credit):

# Example Domain
…
200 fetch 1

Features

• 5 output formats from one call: HTML, Markdown, plain text, PDF, or a JSON envelope with extracted data.
• JavaScript rendering with js_render: true, a real Chromium with engine: "chromium", or mode: "auto", which escalates from a plain fetch only when needed and bills only the step that worked.
• Structured data extraction with CSS selectors, XPath, a typed JSON Schema, or page metadata (JSON-LD, OpenGraph), at no extra credits.
• Screenshots and PDFs of a full page or a single element.
• Batch jobs of up to 10,000 URLs per call (100,000 per open job), with results streamed as JSON Lines.
• Sessions that keep cookies and storage across scrapes, and your own proxy (HTTP, HTTPS, SOCKS5) on any request.
• Billing-safe retries: a POST is retried only when the server provably did no work, so a retry never bills a scrape twice.
• Zero runtime dependencies, ESM and CommonJS builds, and types generated from the OpenAPI spec.

Scrape a web page

Every snippet below runs as an ES module with this setup:

import { Spicrawl } from "@spicrawl/sdk";
const spicrawl = new Spicrawl();

scrape(): the JSON envelope

scrape(params, options?) sends POST /v1/scrape and returns the typed JSON envelope: status (the target site's status), content, headers, credits, engine, warnings, plus data, links or screenshots when you ask for them.

const page = await spicrawl.scrape({ url: "https://example.com", js_render: true, wait_for: "h1" });
console.log(page.status, page.engine, page.credits);
console.log(page.content.slice(0, 200));

A site answering 404 or 403 is a successful call, not an error. The SDK throws only when Spicrawl itself fails.

scrapeRaw(): the document as served

scrapeRaw(params, options?) returns the body itself, discriminated by format:

| format | body | When | |---|---|---| | html, markdown, text | string | response_format is html (default), markdown or text | | pdf | ArrayBuffer | response_format: "pdf" with engine: "chromium" | | json | string, plus a parsed envelope | json, or any of extract, links, screenshot, autoparse, network_capture |

const r = await spicrawl.scrapeRaw({ url: "https://example.com", response_format: "text" });
if (r.format === "text") console.log(r.body);

Extract structured data from a web page

Map field names to CSS selectors, or send a JSON Schema whose properties carry a selector to get typed, validated values:

const product = await spicrawl.scrape({
  url: "https://example.com",
  extract: {
    type: "object",
    properties: {
      title: { type: "string", selector: "h1" },
      summary: { type: "string", selector: "p" },
    },
    required: ["title"],
  },
});
console.log(product.data, product.empty_fields); // empty_fields: selectors that matched nothing

• extract: { title: "h1", price: ".price" } is the short selector form.
• autoparse: true returns JSON-LD, OpenGraph and microdata.
• links: true returns every link on the page in links.
• ai_extract (extraction from a plain-language prompt) is typed but coming soon: the API refuses it today at 0 credits. See the AI extraction guide.

Take a screenshot or save a page as PDF

Screenshots and PDFs need a browser that can draw the page, so set engine: "chromium". Images arrive base64-encoded in screenshots:

import { writeFile } from "node:fs/promises";

const shot = await spicrawl.scrape({
  url: "https://example.com",
  engine: "chromium",
  screenshot: true,
  screenshot_fullpage: true, // or screenshot_selector: "#chart"
});
const img = shot.screenshots?.[0];
if (img) await writeFile("page.png", Buffer.from(img.data, "base64"));

const pdf = await spicrawl.scrapeRaw({ url: "https://example.com", engine: "chromium", response_format: "pdf" });
if (pdf.format === "pdf") await writeFile("page.pdf", Buffer.from(pdf.body));

Batch scrape thousands of URLs

A batch job queues up to 10,000 URLs per call and runs them server-side with retries. Each item is billed only if it succeeds. Batch jobs never use the result cache.

batch.create(), batch.waitForCompletion() and batch.results()

const job = await spicrawl.batch.create({
  urls: ["https://example.com", "https://example.org"],
  js_render: true,
});

const done = await spicrawl.batch.waitForCompletion(job.id, {
  pollIntervalMs: 3000,        // default 3000
  timeoutMs: 30 * 60_000,      // default: wait forever
  onProgress: (j) => console.log(j.status, j.progress.percent_complete),
});

for await (const line of spicrawl.batch.results(job.id)) {
  console.log(line.seq, line.status, line.url, line.http_status ?? line.error?.code);
}

create() is never retried once it may have reached the server, because resubmitting creates and bills a second job. results() follows X-Next-Cursor across pages; on a running job it stops at the last finished item. Per-item overrides go in items: [{ url, external_id, js_render }] instead of urls.

Other batch methods

const one = await spicrawl.batch.get(job.id);                    // a job and its live progress
for await (const j of spicrawl.batch.list({ status: "running" })) console.log(j.id);
await spicrawl.batch.cancel(job.id);                             // -> cancelling, then cancelled
const again = await spicrawl.batch.retry(job.id);                // re-run every failed item

const open = await spicrawl.batch.create({ urls: ["https://example.com"], open: true });
await spicrawl.batch.append(open.id, { urls: ["https://example.org"] }); // not idempotent
await spicrawl.batch.close(open.id);                             // an open job never finishes on its own

const page1 = await spicrawl.batch.resultsPage(job.id, { limit: 100, status: "failed" }); // one page: { lines, nextCursor }
const item = await spicrawl.batch.taskContent(job.id, 0);        // one item's payload: { content, truncated }

Keep cookies and logins with sessions

A session keeps cookies and web storage across scrapes, so a login carries over to later requests.

const session = await spicrawl.sessions.create();                // sessions.create(body?)
await spicrawl.scrape({ url: "https://example.com", session_id: session.id, js_render: true });

const same = await spicrawl.sessions.get(session.id);
for await (const s of spicrawl.sessions.list({ limit: 50 })) console.log(s.id);
const ctx = await spicrawl.sessions.context(session.id);         // cookies + storage: contains credentials, never log it
await spicrawl.sessions.release(session.id);                     // stops accepting work, purges context
await spicrawl.sessions.delete(session.id, { force: true });     // force: even while a task holds it

Request history and credit usage

const recent = await spicrawl.requests.list({ only_errors: "true", limit: 20 });
const trace = await spicrawl.requests.get(recent.data[0].id);    // or an error's requestId

const summary = await spicrawl.usage.summary();                  // plan, credits, allowance, current period
const byDay = await spicrawl.usage.get({ from: "2026-09-01", to: "2026-09-30", group_by: "day" });
const recon = await spicrawl.usage.reconciliation({ day: "2026-09-30" });

requests.list({ all_projects: "true" }) covers every project in the organization and needs a key with the read scope.

Remote browser (coming soon)

spicrawl.browser.token() and spicrawl.browser.connectURL() are typed for the remote CDP browser, which is not available yet. They exist so your code will not change when it launches. See the CDP browser guide.

Pagination

batch.list(), sessions.list() and requests.list() return a PagePromise. await it for one page, or for await it to walk every item on every page:

let page = await spicrawl.requests.list({ limit: 50 });
console.log(page.data.length, page.hasNextPage());
if (page.hasNextPage()) page = await page.getNextPage();
for await (const p of page.iterPages()) console.log(p.data.length);

Response metadata: credits, rate limits, request ids

Every result carries a typed, non-enumerable _meta read from the response headers, so it stays out of JSON.stringify and spreads:

| Field | Header | Meaning | |---|---|---| | requestId | X-Request-Id | pass to requests.get() for the full trace | | creditsCharged | X-Credits-Charged | what this call billed | | creditsRemaining | X-Credits-Remaining | monthly allowance left after it; null when not sent | | requestCost | X-Request-Cost | price of the engine that ran | | engine | X-Engine | engine that produced the page | | cacheState | Cache-State | hit, miss or bypass | | targetStatus | X-Target-Status | the target site's status | | rateLimit | X-RateLimit-* | { limit, remaining, reset } | | warnings | X-Warning | non-fatal CODE: message strings |

Error handling

Every error the SDK throws is a SpicrawlError. HTTP errors map to subclasses by status:

| Class | Status | Meaning | |---|---|---| | BadRequestError | 400, and any status without its own class | invalid or incompatible parameters | | AuthenticationError | 401 | missing, invalid, revoked or expired key | | InsufficientCreditsError | 402 | the monthly credit allowance is used up | | PermissionDeniedError | 403 | missing scope, or engine not entitled | | NotFoundError | 404 | unknown job, session or request | | ConflictError | 409 | e.g. a busy session, or a job already finished | | GoneError | 410 | beyond retention, or a released session | | RateLimitError | 429 | rate, concurrency or live-session limit | | InternalServerError | 5xx | platform failure | | APIConnectionError | none (status 0) | DNS failure, refused or reset connection | | APITimeoutError | none (status 0) | timeoutMs exceeded; subclass of APIConnectionError |

Each error carries status, code (a stable ERR::FAMILY::NAME), retryable (the server's verdict), retryAfterSeconds, requestId, diagnostics.hint and the raw problem body. ERROR_CODES lists all 48 documented codes.

import { Spicrawl, RateLimitError, isSpicrawlError } from "@spicrawl/sdk";

const spicrawl = new Spicrawl();
try {
  await spicrawl.scrape({ url: "https://example.com" });
} catch (err) {
  if (err instanceof RateLimitError) console.log("retry in", err.retryAfterSeconds, "s");
  else if (isSpicrawlError(err)) console.log(err.code, err.retryable, err.diagnostics?.hint, err.requestId);
  else throw err;
}

Retries and timeouts

The API has no idempotency keys, so a repeated scrape is a second, billed scrape. The SDK retries accordingly:

• maxRetries defaults to 2 retries after the first attempt.
• Backoff is exponential with jitter: about 0.5 s, 1 s, 2 s, capped at 8 s.
• A Retry-After header (or retry_after_seconds in the body) is honoured up to 60 s. A longer wait is thrown as the error instead.
• A server answer of retryable: false is never retried.

| Failure | GET, DELETE | POST (scrape, batch, sessions) | |---|---|---| | Connection error | retried | retried only if the connection never opened | | Timeout | retried | never: the server may still be working and billing | | 408, 5xx | retried | never | | 429 | retried | retried: it is refused before any work | | Other 4xx | never | never |

timeoutMs is per attempt and defaults to 180,000 ms (3 minutes, because browser renders can be slow). An AbortSignal cancels the request and any pending retry sleep. Override any of these per client or per call:

const strict = new Spicrawl({ maxRetries: 0, timeoutMs: 60_000 });
await spicrawl.scrape(
  { url: "https://example.com" },
  { timeoutMs: 30_000, maxRetries: 0, signal: AbortSignal.timeout(45_000), headers: { "X-Trace": "abc" } },
);

Configuration

| Option | Environment variable | Default | |---|---|---| | apiKey | SPICRAWL_API_KEY | required; the constructor throws without one | | baseURL | SPICRAWL_BASE_URL | https://api.spicrawl.com | | timeoutMs | none | 180000, per attempt | | maxRetries | none | 2 | | fetch | none | globalThis.fetch (pass your own for a proxy agent or tests) | | defaultHeaders | none | {}, sent with every request |

Runtime support

| Runtime | Status | |---|---| | Node.js 18 and later | Supported (engines: { node: ">=18" }); CI runs Node.js 22 | | ESM (import) | dist/index.js with index.d.ts | | CommonJS (require) | dist/index.cjs with index.d.cts | | Bun, Deno, edge runtimes | Not tested. The SDK needs only the standard fetch, AbortController, Headers and Response; pass apiKey where there is no process.env | | Browsers | Not supported: never ship an API key to a browser. Call Spicrawl from your server |

const { Spicrawl } = require("@spicrawl/sdk"); // CommonJS

Use with AI agents

Give a model a web-reading tool by wrapping one SDK call. This uses the Anthropic SDK (npm install @anthropic-ai/sdk) with a manual tool loop:

import Anthropic from "@anthropic-ai/sdk";
import { Spicrawl } from "@spicrawl/sdk";

const spicrawl = new Spicrawl();
const anthropic = new Anthropic();

const tools: Anthropic.Tool[] = [{
  name: "fetch_page",
  description: "Fetch a web page and return its main content as Markdown.",
  input_schema: { type: "object", properties: { url: { type: "string" } }, required: ["url"] },
}];

async function fetchPage(url: string): Promise<string> {
  const page = await spicrawl.scrapeRaw({ url, response_format: "markdown", main_content_only: true });
  return page.format === "markdown" ? page.body : "";
}

const messages: Anthropic.MessageParam[] = [{ role: "user", content: "Summarise https://example.com" }];
for (;;) {
  const res = await anthropic.messages.create({ model: "claude-opus-5-5", max_tokens: 16000, tools, messages });
  messages.push({ role: "assistant", content: res.content });
  if (res.stop_reason !== "tool_use") break;
  const results: Anthropic.ToolResultBlockParam[] = [];
  for (const block of res.content) {
    if (block.type !== "tool_use") continue;
    const { url } = block.input as { url: string };
    results.push({ type: "tool_result", tool_use_id: block.id, content: await fetchPage(url) });
  }
  messages.push({ role: "user", content: results });
}

With the OpenAI SDK, pass the same function as a function tool and answer each tool_calls entry with a tool message:

import OpenAI from "openai";

const openai = new OpenAI();
const completion = await openai.chat.completions.create({
  model: "gpt-5",
  messages: [{ role: "user", content: "Summarise https://example.com" }],
  tools: [{
    type: "function",
    function: {
      name: "fetch_page",
      description: "Fetch a web page and return its main content as Markdown.",
      parameters: { type: "object", properties: { url: { type: "string" } }, required: ["url"] },
    },
  }],
});
for (const call of completion.choices[0].message.tool_calls ?? []) {
  if (call.type !== "function") continue;
  const markdown = await fetchPage(JSON.parse(call.function.arguments).url); // fetchPage from above
  // send back { role: "tool", tool_call_id: call.id, content: markdown }, then call the model again
}

SDK vs CLI vs MCP vs REST API

• If you write Node.js or TypeScript code, use this SDK: typed requests and errors, pagination, batch helpers and billing-safe retries.
• If you work in a terminal, a shell script or CI, use the Spicrawl CLI (npm install -g @spicrawl/cli).
• If an AI agent (Claude Code, Cursor, Codex, VS Code) should decide what to fetch, connect it to the hosted MCP server at https://mcp.spicrawl.com/mcp with Authorization: Bearer <API key>. See the MCP guide, @spicrawl/mcp and OfficialSpicrawl/agent-plugins.
• If you use another language, call the REST API directly. See the API reference.

FAQ

Is Spicrawl free?

Spicrawl bills credits per successful request, and every organization gets a monthly credit allowance: during the beta, one plan of 1,000 credits per month. A plain fetch costs 1 credit, js_render 3 and engine: "chromium" 8; failed requests cost 0. See Credits.

Are cached results free?

No. A cache hit is billed at the same price as the fetch that stored it; the cache saves latency and load on the target site, not credits. Send cache: false for a fresh fetch at no extra cost. See Caching.

Does it render JavaScript?

Yes. Set js_render: true for a JavaScript-capable engine, engine: "chromium" for a real Chromium, or mode: "auto" to escalate from a plain fetch only when needed; wait_for waits for a CSS selector.

Why didn't my scrape throw when the site returned 404?

The site's status is data, not an error. Read it from status in the envelope, targetStatus in raw mode, or _meta.targetStatus.

Can a retry bill me twice?

Not from the SDK's own retries. POST requests are retried only on 429 or when the connection never opened, both cases where no work was done. Your own retry of a scrape that succeeded is billed again.

Can I crawl a whole website?

Not with one call: there is no site-crawl or sitemap endpoint. Collect URLs with links: true, then submit them as a batch job.

Does it work with Claude Code or Cursor?

The SDK is for your own code. For Claude Code, Cursor, Codex or VS Code, use the hosted MCP server, or run npx @spicrawl/cli init to configure them. See the MCP guide.

Is there a hosted MCP server?

Yes, at https://mcp.spicrawl.com/mcp, authenticated with Authorization: Bearer <API key>.

Is the SDK open source?

Yes, under Apache-2.0 at github.com/OfficialSpicrawl/sdk. The Spicrawl API it calls is a hosted service.

Links

• Documentation: https://docs.spicrawl.com, with llms.txt for AI agents
• Dashboard and API keys: https://app.spicrawl.com
• Runnable examples: scrape-to-markdown.ts, extract-product.ts, batch-crawl.ts
• Bugs and feature requests: https://github.com/OfficialSpicrawl/sdk/issues

License

Apache-2.0