# Firecrawl MCP — Web Search & Scrape Tools

> Add Firecrawl MCP to your agent to search, scrape, and extract full-page content. Run via npx with an API key; fits Cursor, Claude Code, VS Code.

## Install

Merge the JSON below into your `.mcp.json`:

# Firecrawl MCP — Web Search & Scrape Tools

> Add Firecrawl MCP to your agent to search, scrape, and extract full-page content. Run via npx with an API key; fits Cursor, Claude Code, VS Code.

## Quick Use

1) Install

```bash
export FIRECRAWL_API_KEY=fc-YOUR_API_KEY && npx -y firecrawl-mcp
```

2) Run

```bash
Ask your agent: "Search the web for X and summarize the key points with sources."
```

3) Verify

```bash
Confirm the MCP server starts and your client lists Firecrawl tools
```

---

## Intro

**Best for**: agents that need reliable web data (search + scraping) and want a single MCP tool surface instead of ad-hoc curl + copy/paste

**Works with**: MCP-capable clients (Cursor, Claude Code, VS Code), Node.js + npx, Firecrawl API key

**Setup time**: 6 minutes

### Quant Data

- Runs via `npx -y firecrawl-mcp` (repo)
- Setup time ~6 minutes

---

## How to Use It Well

Start by using Firecrawl for one concrete task: search → scrape 3–5 URLs → extract structured bullets. Once the output is stable, formalize it as a reusable agent instruction.

### Adoption Checklist

- Start with one real task and keep the scope narrow
- Capture a baseline: time-to-first-success and output quality
- Version your config/skills so teammates stay in sync

### Guardrails

Keep prompts strict: always ask for sources and limit pages. Web tools can explode context and cost if you let the agent wander.

### FAQ

**Q: Do I need self-hosting?**
A: No. The default mode uses the cloud API with `FIRECRAWL_API_KEY`. Self-hosting is optional if you run your own endpoint.

**Q: Which clients does it work with?**
A: Any MCP client that can run stdio command servers; the repo includes Cursor and VS Code configuration examples.

**Q: How do I control cost?**
A: Use the repo's credit monitoring env vars and keep tasks narrow (domain + query + output format) to reduce retries and page volume.

---

## Source & Thanks

> GitHub: https://github.com/firecrawl/firecrawl-mcp-server
> Owner avatar: https://avatars.githubusercontent.com/u/135057108?v=4
> License (SPDX): MIT
> GitHub stars (verified via `api.github.com/repos/firecrawl/firecrawl-mcp-server`): 6,276


---

<!-- ZH -->

# Firecrawl MCP——网页搜索与抓取工具

> 把 Firecrawl MCP 接入 Claude Code/Cursor：用 API Key 即可做网页搜索、抓取、结构化提取与自动化操作。npx 一条命令启动，适合调研、比价、资料汇总。

## 快速使用

1）安装

```bash
export FIRECRAWL_API_KEY=fc-YOUR_API_KEY && npx -y firecrawl-mcp
```

2）运行

```bash
Ask your agent: "Search the web for X and summarize the key points with sources."
```

3）验证

```bash
Confirm the MCP server starts and your client lists Firecrawl tools
```

---

## 简介

**适合谁**：需要可靠网页数据（搜索+抓取）的 AI 代理/团队，希望用一个 MCP 工具面替代零散的 curl + 复制粘贴

**适用环境**：支持 MCP 的客户端（Cursor、Claude Code、VS Code 等）、Node.js + npx、Firecrawl API Key

**安装耗时**：6 分钟

### 量化信息

- 可用 `npx -y firecrawl-mcp` 启动（仓库）
- 装机约 6 分钟

---

## 用好它的方式

先用 Firecrawl 完成一个明确任务：搜索 → 抓取 3–5 个 URL → 输出结构化要点。输出稳定后，再把它固化为可复用的 agent 指令。

### 推广清单

- 先选一个真实任务，小范围试跑
- 记录基线：首次成功耗时与输出质量
- 配置/技能要版本化，避免团队漂移

### 风险与护栏

提示词要严格：要求带来源、限制页数。网页工具一旦放任探索，很容易造成上下文与成本爆炸。

### FAQ

**Q: 必须自建吗？**
A: 不必。默认用云 API + `FIRECRAWL_API_KEY` 即可运行；只有在你需要自定义端点时才考虑自建。

**Q: 哪些客户端能用？**
A: 只要是能运行 stdio 命令服务器的 MCP 客户端都可以；仓库提供了 Cursor/VS Code 的配置示例。

**Q: 怎么控制费用？**
A: 用仓库的 credit 监控环境变量，并把任务拆小（域名+查询+输出格式），降低重试和抓取页数。

---

## 来源与感谢

> GitHub：https://github.com/firecrawl/firecrawl-mcp-server
> Owner avatar：https://avatars.githubusercontent.com/u/135057108?v=4
> 许可证（SPDX）：MIT
> GitHub stars（已通过 `api.github.com/repos/firecrawl/firecrawl-mcp-server` 核验）：6,276


---
Source: https://tokrepo.com/en/workflows/firecrawl-mcp-web-search-scrape-tools
Author: MCP Hub