# Web scraping MCP Claude Code

Web scraping MCP Claude Code: we plugged Firecrawl's hosted MCP and the FATSEARCH MCP into Claude Code and ran the same 51 web tasks once each. FATSEARCH got about the same right answers with 7.6× fewer tokens, at 3.6× lower cost. On its slowest runs it was slower.

> We make FATSEARCH. The numbers below are our own runs, dated. Where another reader is the better pick, we say so.

## Results

| | Firecrawl MCP | FATSEARCH MCP |
|---|---|---|
| Right answers, of 51 | 51/51 | 50/51 |
| Tokens per task, median | 38,129 | 5,026 |
| Cost per thousand tasks | $53.48 | $15.06 |
| Time per task, median / slowest tenth | 6.7 s / 21.5 s | 5.3 s / 43.8 s |

<small class="note">Claude Code 2.1.280, claude-sonnet-5, headless, built-in WebFetch and WebSearch off, one MCP server per side. Right = the final answer matched the known one; right answers also found in a tool result: Firecrawl 51/51, FATSEARCH 50/51. Tokens = Claude Code's own usage count. Cost = the Claude bill at list price plus the tool: Firecrawl credits at the Hobby rate ($16 per 5,000 credits), FATSEARCH = the Oxylabs price as `fats` prints it.</small>

## Speed: the same on most tasks, slower on blocked pages

A typical task took about as long on both: FATSEARCH 5.3 s, Firecrawl 6.7 s. The slowest runs took FATSEARCH 43.8 s and Firecrawl 21.5 s.

<small class="note">Slowest runs = the slowest tenth. FATSEARCH had 5 runs over a minute, the worst 191 s with 31 MCP calls. On blocked pages it waits for the paid render, and a known retry bug makes the agent wait again; we are fixing it. Firecrawl's server cache already held most pages; FATSEARCH started empty, so Firecrawl's times are a best case.</small>

## Why the token gap

Most of it is Firecrawl's tool list. Claude Code sends it with every model call: 10,977 extra tokens per call.

<small class="note">Firecrawl registers 16 tools, FATSEARCH 1. The first model call, before any page, is 12,044 tokens with Firecrawl and 1,069 with FATSEARCH; that difference, resent on every call, is 87 % of the input-token gap. The rest is page size: 10 Firecrawl runs returned more than Claude Code keeps in context, so it saved the result to a file and the agent read it back.</small>

## Where Firecrawl is better

- **Spotify.** The one task only Firecrawl got right. FATSEARCH got a bare JS shell, and Oxylabs' render failed.
- **Search rescue.** When its scrape refused a site (Reddit, Instagram, Yelp, the New York Times), the agent searched with Firecrawl and found the answer in a result snippet.
- **Speed on blocked pages**, see above.

## Firecrawl's paid lookups

On LinkedIn the agent used Firecrawl's paid data lookups. That one run cost 57 credits, 50 of them from one call.

<small class="note">Firecrawl calls these Alexandria enrichment; the costly call went to people-data-labs. All runs together: 203 credits ($0.65 at Hobby). FATSEARCH's Oxylabs bill: $0.025.</small>

## Setup

Firecrawl, from their [Claude Code quickstart](https://docs.firecrawl.dev/quickstarts/claude-code):

```sh
claude mcp add --transport http firecrawl https://mcp.firecrawl.dev/v2/mcp-oauth
```

<small class="note">Then run `/mcp` in Claude Code, pick firecrawl and sign in in the browser. Without an account, a keyless mode covers scrape, search and parse. Checked 2026-10-19.</small>

FATSEARCH, from the [installation guide](/docs/installation/#claude-code):

```sh
claude mcp add --scope user fats -- fats mcp
```

<small class="note">A local stdio server with 1 tool. Reading pages needs no key; search needs an Oxylabs or Bright Data key ([key setup](/docs/installation/#4-keys-optional)). Full setup with checks: [web search MCP for Claude Code](/guides/web-search-mcp-for-claude-code/).</small>

## Method

One run per side on the same tasks, 2026-10-19, from one machine in Lithuania. We ran FATSEARCH twice more on the same tasks to check stability: 99/102 right.

<small class="note">The scorecard with every number: [benchmarks](/benchmarks/#claude-code). One page through every reader: [Firecrawl alternatives](/compare/firecrawl-alternatives/) and the [comparison hub](/compare/). We make FATSEARCH; these are our runs.</small>

## Questions

### Can I use Firecrawl with Claude Code?

Yes: run claude mcp add --transport http firecrawl https://mcp.firecrawl.dev/v2/mcp-oauth, then sign in through /mcp. In our 51-task run it got 51/51 right.

### Why is my Claude code using so many tokens?

Every MCP server's tool list is sent with every model call: Firecrawl's added 10,977 tokens per call over FATSEARCH's single tool. Fewer tools and smaller page output mean fewer tokens per task.

### Is Firecrawl MCP free?

Partly: a keyless mode covers scrape, search and parse, and a free-plan key comes with a monthly credit allowance. Paid use starts at the Hobby plan, $16 per 5,000 credits.

## Related

- [Web scraping MCP: every reader we measured](https://fatsearch.com/compare/)
- [Benchmarks: every run, charted](https://fatsearch.com/benchmarks/)
- [Method: one page, six readers](https://fatsearch.com/docs/benchmarks/#one-page-six-readers)
- [Method: the hard-pages run](https://fatsearch.com/docs/benchmarks/#other-readers)

By FATSEARCH · data as of 2026-10-19 · reviewed 2026-10-20 · https://fatsearch.com/guides/web-scraping-mcp-claude-code/