Document Parsing

Updated Feb 2026

Firecrawl automatically detects and converts documents (PDF, DOCX, XLSX) into clean, structured markdown. Just pass a document URL to the scrape endpoint.

Direct File Upload

For local files not accessible by URL, use the dedicated Parse endpoint — it accepts file uploads directly via multipart/form-data and is up to 5x faster.

Supported Formats

Format	Extensions	Output
PDF	.pdf	Extracted text with layout preservation; OCR for scanned docs
Word	.docx, .doc, .odt, .rtf	Headings, paragraphs, lists, tables preserved
Excel	.xlsx, .xls	Worksheets as HTML tables, sheet names as H2 headings

How It Works

Document parsing is built into the standard scrape endpoint. Firecrawl detects the file type via URL extension or content-type header and automatically processes it.

Basic Usage

Python

from firecrawl import Firecrawl

firecrawl = Firecrawl(api_key="fc-YOUR-API-KEY")

# PDF document
result = firecrawl.scrape("https://example.com/report.pdf")
print(result['markdown'])

# Word document
result = firecrawl.scrape("https://example.com/proposal.docx")
print(result['markdown'])

# Excel spreadsheet
result = firecrawl.scrape("https://example.com/data.xlsx")
print(result['markdown'])

Node.js

import Firecrawl from '@mendable/firecrawl-js';

const firecrawl = new Firecrawl({ apiKey: "fc-YOUR-API-KEY" });

// PDF document
const doc = await firecrawl.scrape("https://example.com/report.pdf", {
  formats: ["markdown"]
});
console.log(doc.markdown);

// Word document
const wordDoc = await firecrawl.scrape("https://example.com/proposal.docx", {
  formats: ["markdown"]
});
console.log(wordDoc.markdown);

// Excel spreadsheet
const excelDoc = await firecrawl.scrape("https://example.com/data.xlsx", {
  formats: ["markdown"]
});
console.log(excelDoc.markdown);

cURL

curl -X POST https://api.firecrawl.dev/v2/scrape \
  -H 'Content-Type: application/json' \
  -H 'Authorization: Bearer fc-YOUR-API-KEY' \
  -d '{
    "url": "https://example.com/report.pdf",
    "formats": ["markdown"]
  }'

PDF Parsing Modes

Control how PDFs are processed using the parsers option:

Mode	Description	Speed	Use When
auto	Attempts fast text extraction first, falls back to OCR if needed	Fast (with fallback)	Default -- works for most PDFs
fast	Text-only extraction, skips scanned/image-heavy pages	Fastest	Digital-native PDFs with selectable text
ocr	Forces OCR on every page	Slowest	Scanned documents, image-heavy PDFs

Configuring PDF Parsers

# Auto mode (default)
result = firecrawl.scrape(
    "https://example.com/report.pdf",
    formats=["markdown"]
)

# Force OCR for a scanned document
result = firecrawl.scrape(
    "https://example.com/scanned-contract.pdf",
    formats=["markdown"],
    parsers=[{"type": "pdf", "mode": "ocr"}]
)

# Fast mode for text-based PDFs
result = firecrawl.scrape(
    "https://example.com/digital-report.pdf",
    formats=["markdown"],
    parsers=[{"type": "pdf", "mode": "fast"}]
)

# Limit page count
result = firecrawl.scrape(
    "https://example.com/long-report.pdf",
    formats=["markdown"],
    parsers=[{"type": "pdf", "mode": "auto", "maxPages": 20}]
)

// Force OCR
const doc = await firecrawl.scrape("https://example.com/scanned.pdf", {
  formats: ["markdown"],
  parsers: [{ type: "pdf", mode: "ocr" }]
});

// Fast mode with page limit
const doc = await firecrawl.scrape("https://example.com/report.pdf", {
  formats: ["markdown"],
  parsers: [{ type: "pdf", mode: "fast", maxPages: 50 }]
});

curl -X POST https://api.firecrawl.dev/v2/scrape \
  -H 'Content-Type: application/json' \
  -H 'Authorization: Bearer fc-YOUR-API-KEY' \
  -d '{
    "url": "https://example.com/scanned.pdf",
    "formats": ["markdown"],
    "parsers": [{"type": "pdf", "mode": "ocr", "maxPages": 20}]
  }'

Parser Parameters

Parameter	Type	Default	Description
type	string	Required	Document type ("pdf")
mode	string	"auto"	Parsing mode: auto, fast, ocr
maxPages	number	--	Maximum pages to process

Word Document Output

Word documents preserve their structural elements:

Source Element	Markdown Output
Headings	#, ##, ### (matching level)
Paragraphs	Standard text blocks
Bulleted lists	- Item
Numbered lists	1. Item
Tables	Markdown tables
Bold/italic	**bold** / *italic*

Example Output

# Quarterly Report Q4 2025

## Executive Summary

Revenue increased **15%** year-over-year, driven by:

- New enterprise contracts
- Expansion in APAC region
- Product line additions

## Financial Details

| Metric | Q3 2025 | Q4 2025 | Change |
|--------|---------|---------|--------|
| Revenue | $12.5M | $14.4M | +15% |
| Users | 50,000 | 62,000 | +24% |

Excel Document Output

Each worksheet becomes a section with the sheet name as a heading:

Example Output

## Sheet: Revenue Data

| Month | Revenue | Growth |
|-------|---------|--------|
| January | $1.2M | 5% |
| February | $1.3M | 8% |
| March | $1.5M | 15% |

## Sheet: User Metrics

| Month | Active Users | Churn |
|-------|-------------|-------|
| January | 45,000 | 2.1% |
| February | 48,000 | 1.8% |

JSON Extraction from Documents

Combine document parsing with JSON extraction to pull structured data from documents:

from pydantic import BaseModel
from typing import List

class InvoiceItem(BaseModel):
    description: str
    quantity: int
    unit_price: float
    total: float

class Invoice(BaseModel):
    invoice_number: str
    date: str
    vendor: str
    items: List[InvoiceItem]
    grand_total: float

result = firecrawl.scrape(
    "https://example.com/invoice.pdf",
    formats=[{
        "type": "json",
        "schema": Invoice.model_json_schema()
    }],
    parsers=[{"type": "pdf", "mode": "ocr"}]
)

invoice = result['json']
print(f"Invoice #{invoice['invoice_number']}: ${invoice['grand_total']}")

LlamaParse Integration

For advanced PDF parsing (complex tables, charts, multi-column layouts), configure LlamaParse in self-hosted deployments:

# In .env
LLAMAPARSE_API_KEY=your-llamaparse-key

LlamaParse provides enhanced accuracy for:

• Complex multi-column layouts
• Tables with merged cells
• Charts and diagrams (text extraction)
• Academic papers with equations

Cost

Operation	Credits
Document scrape (base)	1 per page of content
PDF parsing	+1 per PDF page
JSON extraction from document	+4 per page

Cost Examples

Scenario	Credits
10-page PDF, markdown only	11 (1 base + 10 PDF pages)
10-page PDF with JSON extraction	15 (1 base + 10 PDF pages + 4 JSON)
Word document (any length)	1
Excel spreadsheet (any size)	1

Related Pages

• Scrape -- All scrape formats and options
• LLM Extract -- Schema-based extraction details
• Batch Scrape -- Process multiple documents at once
• Pricing -- Full credit cost breakdown

---

Next: Enhanced Mode -->