toolgate-firecrawl-mcp
A recoverable Firecrawl MCP server that adds payment lifecycle, idempotency, and automatic refunds/fallbacks to web scraping, enabling reliable paid scraping via natural language.
README
toolgate-firecrawl-mcp
A recoverable Firecrawl MCP server powered by Toolgate.
Every paid scrape goes through a full lifecycle: balance check → execute → charge → trace. If it fails, you get a refund or fallback — never a silent loss.
What happens when you scrape
Agent calls scrape_url
→ Toolgate checks idempotency (duplicate? return cached)
→ Toolgate checks balance (enough? deduct $0.10)
→ Firecrawl scrapes the URL (JS-rendered, clean markdown)
→ On success: charge confirmed, trace saved
→ On failure: refund issued, trace saved with reason
→ On no balance: fallback (fetch+cheerio metadata) returned free
Quick Start
Install & run (demo mode — no API key needed)
npm install
npm run dev
Demo mode uses fetch+cheerio as the scraping backend. scrape_url and balance tools work out of the box. No Firecrawl credits consumed.
Claude Desktop — demo mode
{
"mcpServers": {
"toolgate-firecrawl": {
"command": "node",
"args": ["/path/to/toolgate-firecrawl-mcp/dist/server.js"],
"env": {
"TOOLGATE_DEMO_MODE": "true"
}
}
}
}
Claude Desktop — production (real Firecrawl)
{
"mcpServers": {
"toolgate-firecrawl": {
"command": "node",
"args": ["/path/to/toolgate-firecrawl-mcp/dist/server.js"],
"env": {
"FIRECRAWL_API_KEY": "fc-your-key-here"
}
}
}
}
Tools
| Tool | Price | Description |
|---|---|---|
scrape_url |
$0.10/call | Scrape a URL → clean markdown. Fallback: basic metadata free. |
web_search |
$0.01/result (min $0.05) | Web search via Firecrawl. Requires API key. |
extract_data |
$0.15/call | Structured JSON extraction with schema. Requires API key. |
check_balance |
free | Check your current Toolgate balance. |
add_balance |
free | Add demo balance (demo mode only). |
view_traces |
free | View recent execution traces with decisions and charge status. |
Demo
npm run scenario
Example output:
{
"passed": 6,
"failed": 0,
"skipped": 2,
"scenarios": [
{ "name": "url_validation_ssrf", "passed": true },
{ "name": "payment_missing_fallback", "passed": true },
{ "name": "credit_balance", "passed": true },
{ "name": "paid_scrape", "passed": true },
{ "name": "duplicate_request", "passed": true },
{ "name": "unreachable_url", "skipped": true, "reason": "network" },
{ "name": "free_tier_not_applicable", "passed": true },
{ "name": "trace_output", "passed": true }
]
}
How Toolgate works here
Toolgate wraps every Firecrawl call with a payment lifecycle. Each call gets an idempotency key — retrying with the same key returns the cached result at no extra cost. If the Firecrawl call fails after payment was deducted, the charge is automatically refunded. Every decision is recorded in a queryable execution trace.
Environment Variables
| Variable | Default | Description |
|---|---|---|
FIRECRAWL_API_KEY |
— | Firecrawl API key. Omit for demo mode. |
FIRECRAWL_API_URL |
https://api.firecrawl.dev |
Firecrawl API base URL. |
TOOLGATE_PUBLISHER_KEY |
tg_firecrawl_demo |
Toolgate publisher key. |
TOOLGATE_DEMO_MODE |
true if no API key |
Force demo mode. |
TOOLGATE_PRELOAD_BALANCE |
1.0 (demo only) |
USD pre-loaded in demo mode. |
SCRAPE_TIMEOUT_MS |
15000 |
Timeout for scrape requests. |
MAX_RESPONSE_BYTES |
5242880 (5MB) |
Max response size. |
Built with
- @tkorkmaz/toolgate — paid-action runtime
- @mendable/firecrawl-js — scraping engine
- @modelcontextprotocol/sdk — MCP server/client
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。