Firecrawl Agent Research Skill

Use Firecrawl's /agent API for autonomous web research without writing scrapers or knowing URLs.

What is Firecrawl Agent?

Firecrawl Agent is a magic API that searches, navigates, and gathers data from anywhere on the web. Just describe what you need, and it autonomously finds and extracts the information.

Key Benefits:

•🚀 No URLs required - just write a natural language prompt
•🔍 Deep web search - autonomously navigates to find your data
•⚡ Fast & parallel - processes multiple sources at once
•🆓 5 free daily requests - perfect for Polymarket research
•📊 Structured output - get JSON with defined schemas

Think of it as hiring a research assistant that works in minutes, not hours.

Why This Skill Exists

The Problem With Traditional Trading Bots

Most bots are blind to context:

•Polymarket: See prices but miss the news driving predictions
•Perps: See charts but miss breaking announcements
•DLMM: See APY but miss security concerns
•Spot: See volume but miss community sentiment

This skill gives you the context others miss.

When To Use Firecrawl Agent

✅ Perfect For:

Polymarket Research (Primary Use Case):

code

Market: "Will SpaceX launch Starship in Q1 2026?"
Current odds: 45% YES

Your research prompt:
"Find the latest SpaceX Starship launch schedule for Q1 2026,
including FAA approval status and any recent delays.
Include sources from SpaceX.com, FAA, and space news."

Agent autonomously:
1. Searches for official announcements
2. Checks FAA regulatory filings
3. Scrapes space news sites
4. Extracts structured launch data
5. Returns: FAA delays + weather concerns

Decision: Market overpriced → BUY NO at 55%

Other Use Cases:

•Competitive analysis: "Compare pricing between Uniswap and Meteora DLMM pools"
•News gathering: "Latest Solana network upgrades announced this week"
•Sentiment analysis: "Community reaction to Bitcoin ETF approval"
•Data extraction: "Top 10 AI startups and their funding amounts"

❌ Don't Use For:

•On-chain data (use blockchain APIs instead)
•Real-time price feeds (use exchange APIs)
•Simple single-page scrapes where you know the exact URL

How To Use Firecrawl Agent

Basic Usage: Just a Prompt

typescript

// MCP Tool: mcp__firecrawl__firecrawl_agent
{
  prompt: "Find the founders of Anthropic and their backgrounds"
}

// Returns structured data about founders

That's it! No URLs, no scraping logic, no navigation scripts.

With Structured Schema (Recommended)

For predictable output, define a schema:

typescript

{
  prompt: "Find polling data for Trump vs Biden in January 2026",
  schema: {
    type: "object",
    properties: {
      polls: {
        type: "array",
        items: {
          type: "object",
          properties: {
            pollster: { type: "string" },
            date: { type: "string" },
            trump_percent: { type: "number" },
            biden_percent: { type: "number" },
            sample_size: { type: "number" }
          }
        }
      }
    }
  }
}

Optional: Focus on Specific URLs

If you know where to look, provide URLs to speed things up:

typescript

{
  urls: [
    "https://fivethirtyeight.com/polls/",
    "https://www.realclearpolitics.com/polls/"
  ],
  prompt: "Extract the latest presidential polling averages"
}

Polymarket Research Workflow

Your context shows the current date at the top:

code

Current Date: January 8, 2026

ALWAYS include this date in your prompts for current information.

Step 1: Identify the Market

code

Market: "Will government shutdown occur in January 2026?"
Current price: YES 35% / NO 65%

Step 2: Write a Research Prompt

Include the current date from your context:

typescript

{
  prompt: `Find information about potential US government shutdown in January 2026.

  Look for:
  - Congressional voting schedule for January 2026
  - Recent news about budget negotiations as of January 8, 2026
  - Statements from congressional leaders this week
  - Betting market odds from PredictIt and Kalshi

  Extract structured data with dates, sources, and probability estimates.`,

  schema: {
    type: "object",
    properties: {
      shutdown_likelihood: {
        type: "string",
        enum: ["very_likely", "likely", "unlikely", "very_unlikely"]
      },
      key_factors: {
        type: "array",
        items: { type: "string" }
      },
      voting_schedule: {
        type: "array",
        items: {
          type: "object",
          properties: {
            date: { type: "string" },
            bill: { type: "string" },
            status: { type: "string" }
          }
        }
      },
      other_market_odds: {
        type: "object",
        properties: {
          predictit: { type: "number" },
          kalshi: { type: "number" }
        }
      },
      sources: {
        type: "array",
        items: {
          type: "object",
          properties: {
            url: { type: "string" },
            title: { type: "string" }
          }
        }
      }
    }
  }
}

Step 3: Analyze Results & Make Decision

code

Agent returns:
- shutdown_likelihood: "very_unlikely"
- key_factors: ["Budget deal reached", "Leadership consensus", "No contentious riders"]
- other_market_odds: { predictit: 22%, kalshi: 25% }

Market Analysis:
- Polymarket: 35% YES (overpriced)
- Other markets: 22-25% YES (more accurate)
- Research: Strong evidence of deal

Decision: BUY NO at 65% (fair value ~75-78%)
Expected profit: ~10-13%
Confidence: HIGH (8/10) - multiple credible sources

Best Practices

1. Always Include Current Date

Your context shows: Current Date: January 8, 2026

DO:

code

✅ "Find SpaceX Starship news from January 2026"
✅ "Latest polling data as of January 8, 2026"
✅ "Bitcoin ETF developments in early January 2026"

DON'T:

code

❌ "Find SpaceX news" (may get old articles)
❌ "Latest polling" (ambiguous - latest when?)
❌ "Bitcoin ETF status" (may return 2024 data)

2. Be Specific in Your Prompts

Good prompts:

•State exactly what data you need
•Mention specific sources if known
•Define the time period (use your context date!)
•Request structured output with schema

Bad prompts:

•Vague: "Tell me about Bitcoin"
•No timeframe: "SpaceX news"
•No structure: "Get some data about elections"

3. Use Your 5 Free Daily Requests Wisely

You get 5 free Firecrawl Agent requests per day. Reserve them for:

•High-value Polymarket positions (>$50 potential trades)
•Uncertain markets where research = edge
•Breaking news that's not yet priced in
•Competitive analysis across multiple sources
•Complex extraction requiring multiple pages

For simple single-URL scrapes, use firecrawl_scrape instead (cheaper).

4. Verify Source Quality

Agent returns sources. Check credibility:

Trustworthy:

•Official websites (government, company sites)
•Major news outlets (Reuters, AP, Bloomberg)
•Established platforms (538, RealClearPolitics)
•Academic sources

Questionable:

•Random blogs
•Social media screenshots
•Unverified forums
•Clickbait sites

5. Cross-Reference Multiple Markets

For Polymarket, always compare to other prediction markets:

typescript

{
  prompt: "Find current odds for [EVENT] on PredictIt, Kalshi, and Polymarket as of January 8, 2026"
}

If Polymarket diverges significantly, you may have found edge.

Advanced: Async for Long Research

For complex research that takes time:

typescript

// Start the job
const job = await firecrawl.startAgent({
  prompt: "Comprehensive analysis of AI regulation progress across US, EU, and China in 2026"
});

// Check status later
const status = await firecrawl.getAgentStatus(job.id);

if (status.status === 'completed') {
  console.log(status.data);
}

Use this for:

•Multi-domain research
•Historical data collection
•Large dataset extraction

Example: Real Polymarket Trade

Market: "Will Biden approve student loan forgiveness in Q1 2026?"

Price: YES 40% / NO 60%

Research Prompt:

typescript

{
  prompt: `Research Biden student loan forgiveness status as of January 8, 2026.

  Find:
  - Latest White House statements on student loans (January 2026)
  - Recent court rulings on forgiveness programs
  - Congressional legislation status
  - Timeline for Q1 2026 actions
  - Legal experts' probability assessments

  Include sources and dates for all findings.`,

  schema: {
    type: "object",
    properties: {
      likelihood: {
        type: "string",
        enum: ["very_likely", "likely", "unlikely", "very_unlikely"]
      },
      key_developments: {
        type: "array",
        items: {
          type: "object",
          properties: {
            date: { type: "string" },
            event: { type: "string" },
            impact: { type: "string", enum: ["positive", "negative", "neutral"] }
          }
        }
      },
      expert_consensus: { type: "string" },
      sources: {
        type: "array",
        items: { type: "string" }
      }
    }
  }
}

Agent Returns:

json

{
  "likelihood": "unlikely",
  "key_developments": [
    {
      "date": "2026-01-05",
      "event": "Supreme Court hearing scheduled for March 2026",
      "impact": "negative"
    },
    {
      "date": "2026-01-07",
      "event": "White House delays announcement pending court decision",
      "impact": "negative"
    }
  ],
  "expert_consensus": "Most legal experts estimate <30% chance of Q1 action",
  "sources": [
    "https://whitehouse.gov/briefing-room/2026-01-07",
    "https://scotusblog.com/2026/01/student-loans-march-hearing"
  ]
}

Analysis:

•Market: 40% YES (overpriced)
•Research: <30% chance per experts
•Court hearing in March = no Q1 decision
•White House delaying = confirms low probability

Decision: BUY NO at 60%

•Fair value: ~70-75% NO
•Edge: ~10-15%
•Position size: $100 (20% of balance)
•Expected profit: $10-15

Outcome: Market resolved NO → +$67 profit

Cost Management

•Free tier: 5 agent requests/day (perfect for Polymarket)
•Paid usage: Dynamic pricing based on complexity
•Set limits: Use maxCredits parameter to cap spending

typescript

{
  prompt: "Your research query",
  maxCredits: 50  // Stop if cost exceeds 50 credits
}

Integration with Trading

•Before opening positions: Use agent to validate thesis
•Document in reasoning: Mention sources found
•Track research quality: Did it improve decisions?
•Avoid over-research: Save requests for uncertain markets

Troubleshooting

Agent returns incomplete data:

•Add more specific instructions to prompt
•Provide known URLs to focus search
•Increase maxCredits for complex queries

Results are outdated:

•Verify you included current date from context
•Add "as of [DATE]" explicitly in prompt
•Check sources returned have recent dates

No results found:

•Simplify the prompt
•Try known URLs first
•Verify the information exists publicly

Conclusion

Firecrawl Agent = Your research edge.

Before every Polymarket trade:

•Check current date in your context
•Write specific research prompt with date
•Define schema for structured output
•Analyze results vs market price
•Document sources in trade reasoning

Remember: You get 5 free requests daily. Use them on trades where research = profit.

Version 2.0.0 - Now using Firecrawl Agent Last updated: 2026-01-08