OpenRouter

Use Firecrawl as a tool with any model served by OpenRouter.

Pair OpenRouter, a unified API for hundreds of LLMs, with Firecrawl to give any model live web search, scrape, and crawl.

OpenRouter's API is OpenAI-compatible, so you can use the OpenAI SDK pointed at OpenRouter's base URL plus Firecrawl's Python or JavaScript SDK as the tool.

Setup

npm install @mendable/firecrawl-js openai zod

export FIRECRAWL_API_KEY=fc-YOUR-API-KEY
export OPENROUTER_API_KEY=sk-or-YOUR-OPENROUTER-KEY

Scrape + Summarize with Any OpenRouter Model

This scrapes a page with Firecrawl and summarizes it with whatever model you pick from OpenRouter, here, Claude Haiku 4.5.

import FirecrawlApp from '@mendable/firecrawl-js';
import OpenAI from 'openai';

const firecrawl = new FirecrawlApp({ apiKey: process.env.FIRECRAWL_API_KEY });

const openrouter = new OpenAI({
apiKey: process.env.OPENROUTER_API_KEY,
baseURL: 'https://openrouter.ai/api/v1',
});

const scraped = await firecrawl.scrape('https://docs.firecrawl.dev', {
formats: ['markdown'],
});

const completion = await openrouter.chat.completions.create({
model: 'anthropic/claude-haiku-4.5',
messages: [
  { role: 'user', content: `Summarize in 5 bullets: ${scraped.markdown}` },
],
});

console.log(completion.choices[0]?.message.content);

Switch the model string to any OpenRouter-supported model, openai/gpt-5, google/gemini-2.5-pro, meta-llama/llama-4-maverick, etc.

Tool Calling: Model Decides When to Scrape

OpenRouter supports OpenAI-style tool calls, so Firecrawl plugs in as a function the model can invoke.

import FirecrawlApp from '@mendable/firecrawl-js';
import OpenAI from 'openai';
import { z } from 'zod';

const firecrawl = new FirecrawlApp({ apiKey: process.env.FIRECRAWL_API_KEY });
const openrouter = new OpenAI({
apiKey: process.env.OPENROUTER_API_KEY,
baseURL: 'https://openrouter.ai/api/v1',
});

const ScrapeArgs = z.object({
url: z.string().describe('The URL to scrape'),
});

const tools = [
{
  type: 'function' as const,
  function: {
    name: 'scrape_website',
    description: 'Scrape the markdown content of any URL via Firecrawl',
    parameters: z.toJSONSchema(ScrapeArgs),
  },
},
];

const response = await openrouter.chat.completions.create({
model: 'anthropic/claude-haiku-4.5',
tools,
messages: [
  {
    role: 'user',
    content: 'What is Firecrawl? Visit firecrawl.dev and tell me.',
  },
],
});

const call = response.choices[0]?.message.tool_calls?.[0];
if (call?.function.name === 'scrape_website') {
const { url } = ScrapeArgs.parse(JSON.parse(call.function.arguments));
const page = await firecrawl.scrape(url, { formats: ['markdown'] });
console.log(page.markdown);
}

Python

import os
from firecrawl import FirecrawlApp
from openai import OpenAI

firecrawl = FirecrawlApp(api_key=os.environ["FIRECRAWL_API_KEY"])

openrouter = OpenAI(
  api_key=os.environ["OPENROUTER_API_KEY"],
  base_url="https://openrouter.ai/api/v1",
)

page = firecrawl.scrape("https://docs.firecrawl.dev", formats=["markdown"])

completion = openrouter.chat.completions.create(
  model="anthropic/claude-haiku-4.5",
  messages=[{"role": "user", "content": f"Summarize: {page.markdown}"}],
)

print(completion.choices[0].message.content)

Notes

• Firecrawl is fully model-agnostic, pick any OpenRouter model without changing the scrape code.
• Many top OpenRouter apps (Cline, Roo Code, Kilo, Cursor, Continue) are themselves agent harnesses that can use Firecrawl MCP, see MCP Server to wire Firecrawl into those directly.
• For large jobs, use batch scrape to stay within LLM context budgets.