Agent Skills: Firecrawl SDK Patterns

Firecrawl SDK Patterns

Overview

Production-ready patterns for Firecrawl SDK (@mendable/firecrawl-js / firecrawl-py). Covers singleton client, typed wrappers, retry with backoff, response validation, and reusable scraping service patterns.

Prerequisites

• @mendable/firecrawl-js installed
• Understanding of async/await patterns
• TypeScript strict mode recommended

Instructions

Step 1: Singleton Client with Configuration

// src/firecrawl/client.ts
import FirecrawlApp from "@mendable/firecrawl-js";

let instance: FirecrawlApp | null = null;

export function getFirecrawl(): FirecrawlApp {
  if (!instance) {
    if (!process.env.FIRECRAWL_API_KEY) {
      throw new Error("FIRECRAWL_API_KEY environment variable is required");
    }
    instance = new FirecrawlApp({
      apiKey: process.env.FIRECRAWL_API_KEY,
      ...(process.env.FIRECRAWL_API_URL
        ? { apiUrl: process.env.FIRECRAWL_API_URL }
        : {}),
    });
  }
  return instance;
}

Step 2: Typed Scrape Wrapper

// src/firecrawl/scrape.ts
import { getFirecrawl } from "./client";

interface ScrapeResult {
  url: string;
  title: string;
  markdown: string;
  links: string[];
  scrapedAt: string;
}

export async function scrapePage(
  url: string,
  options?: { waitFor?: number; includeLinks?: boolean }
): Promise<ScrapeResult> {
  const firecrawl = getFirecrawl();
  const formats: string[] = ["markdown"];
  if (options?.includeLinks) formats.push("links");

  const result = await firecrawl.scrapeUrl(url, {
    formats,
    onlyMainContent: true,
    ...(options?.waitFor ? { waitFor: options.waitFor } : {}),
  });

  if (!result.success) {
    throw new Error(`Scrape failed for ${url}: ${result.error}`);
  }

  return {
    url: result.metadata?.sourceURL || url,
    title: result.metadata?.title || "",
    markdown: result.markdown || "",
    links: result.links || [],
    scrapedAt: new Date().toISOString(),
  };
}

Step 3: Retry with Exponential Backoff

// src/firecrawl/retry.ts
export async function withRetry<T>(
  operation: () => Promise<T>,
  config = { maxRetries: 3, baseDelayMs: 1000, maxDelayMs: 30000 }
): Promise<T> {
  for (let attempt = 0; attempt <= config.maxRetries; attempt++) {
    try {
      return await operation();
    } catch (error: any) {
      if (attempt === config.maxRetries) throw error;

      const status = error.statusCode || error.status;
      // Only retry on rate limits (429) and server errors (5xx)
      if (status && status !== 429 && status < 500) throw error;

      const delay = Math.min(
        config.baseDelayMs * Math.pow(2, attempt) + Math.random() * 500,
        config.maxDelayMs
      );
      console.warn(`Firecrawl retry ${attempt + 1}/${config.maxRetries} in ${delay.toFixed(0)}ms`);
      await new Promise(r => setTimeout(r, delay));
    }
  }
  throw new Error("Unreachable");
}

// Usage: await withRetry(() => scrapePage("https://example.com"))

Step 4: Scraping Service with Queue

// src/firecrawl/service.ts
import PQueue from "p-queue";
import { scrapePage, type ScrapeResult } from "./scrape";
import { withRetry } from "./retry";

export class FirecrawlService {
  private queue: PQueue;

  constructor(concurrency = 3) {
    this.queue = new PQueue({
      concurrency,
      interval: 1000,
      intervalCap: 5,  // max 5 requests per second
    });
  }

  async scrape(url: string): Promise<ScrapeResult> {
    return this.queue.add(() => withRetry(() => scrapePage(url)));
  }

  async scrapeMany(urls: string[]): Promise<ScrapeResult[]> {
    return Promise.all(urls.map(url => this.scrape(url)));
  }

  get pending(): number {
    return this.queue.pending;
  }
}

Step 5: Response Validation with Zod

import { z } from "zod";

const FirecrawlScrapeResponse = z.object({
  success: z.literal(true),
  markdown: z.string().min(1),
  metadata: z.object({
    title: z.string().optional(),
    sourceURL: z.string().url(),
    statusCode: z.number().optional(),
  }),
});

export function validateScrapeResponse(result: unknown) {
  const parsed = FirecrawlScrapeResponse.safeParse(result);
  if (!parsed.success) {
    console.error("Invalid Firecrawl response:", parsed.error.issues);
    return null;
  }
  return parsed.data;
}

Step 6: Python Patterns

# firecrawl_service.py
import os
from firecrawl import FirecrawlApp
from functools import lru_cache
import time

@lru_cache(maxsize=1)
def get_firecrawl() -> FirecrawlApp:
    """Singleton Firecrawl client."""
    return FirecrawlApp(api_key=os.environ["FIRECRAWL_API_KEY"])

def scrape_with_retry(url: str, max_retries: int = 3) -> dict:
    """Scrape with exponential backoff."""
    for attempt in range(max_retries):
        try:
            return get_firecrawl().scrape_url(url, params={
                "formats": ["markdown"],
                "onlyMainContent": True,
            })
        except Exception as e:
            if attempt == max_retries - 1:
                raise
            delay = (2 ** attempt) + (time.time() % 1)
            print(f"Retry {attempt + 1}/{max_retries} in {delay:.1f}s: {e}")
            time.sleep(delay)

Output

• Singleton client with env-based configuration
• Typed wrappers returning clean domain objects
• Automatic retry with exponential backoff + jitter
• Queue-based concurrency control
• Zod validation for response safety

Error Handling

| Pattern | Use Case | Benefit | |---------|----------|---------| | Singleton client | All SDK usage | One instance, consistent config | | Typed wrapper | Business logic | Compile-time safety | | Retry + backoff | 429 / 5xx errors | Automatic recovery | | Queue | Multiple URLs | Respect rate limits | | Zod validation | Any API response | Catch API changes early |

Examples

Factory Pattern (Multi-Tenant)

const clients = new Map<string, FirecrawlApp>();

export function getClientForTenant(tenantId: string): FirecrawlApp {
  if (!clients.has(tenantId)) {
    const apiKey = getTenantApiKey(tenantId);
    clients.set(tenantId, new FirecrawlApp({ apiKey }));
  }
  return clients.get(tenantId)!;
}

Resources

• Node SDK
• Python SDK
• p-queue
• Zod

Next Steps

Apply patterns in firecrawl-core-workflow-a for real-world usage.