# scrapi [Health: Active]

**Category:** 🔎 Search & Data Extraction  
**Repository:** https://github.com/bamchi/scrapi-mcp-server  
**GitHub Stars:** 5  
**npm Downloads (last month):** 171  
**Views:** 1  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/scrapi

## Description
Web scraping for AI agents. Converts URLs to clean, LLM-ready Markdown with anti-bot bypass.

## Tools
Capabilities this server exposes over MCP:

- **scrape_url** — Scrapes a webpage and returns the content in AI-readable Markdown format. Can access blocked sites through browser rendering.
- **scrape_urls** — Scrapes multiple webpages in parallel and returns the content in AI-readable Markdown format. Can access blocked sites through browser rendering.
- **get_usage** — Check API usage and remaining credits. Returns current plan, subscription credits, purchased credits, and total remaining credits.
- **scraper_server_status** — Check the status of all ScraperServer instances. Shows server health, circuit breaker state, failure counts, and last success/failure times.
- **get_billing** — Retrieve MCP billing information: subscription details, available plans, daily usage history, or spending limits.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "scrapi": {
    "command": "npx",
    "args": ["-y","@scrapi.ai/mcp-server"],
    "env": {
      "SCRAPI_API_KEY": ""
    }
  }
}
```

**Requires environment variables:** `SCRAPI_API_KEY` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What scrapi MCP server does

The scrapi MCP server gives an MCP client a way to fetch readable content from a web URL. It sends the requested page to Scrapi and returns content prepared for language-model processing rather than raw page HTML. Markdown output can preserve headings, lists, and links, while text output provides a plain-text version.

The service also identifies page metadata such as title, author, publication date, and images when available. Its documented processing removes advertisements, navigation, and scripts. JavaScript rendering is supported for single-page applications, and the service includes automatic retries for HTTP 429 responses using the server-provided Retry-After value.

## How it works

The exposed MCP operation is `scrape_url`. It requires a `url` string and accepts an optional `format` value. The format defaults to `markdown`; `text` selects plain-text extraction. A typical request supplies a public article or page URL and asks the agent to use the returned content for tasks such as summarization or analysis.

The server can run over stdio or Streamable HTTP. Stdio starts the package through `npx` or a locally built Node.js entry point. HTTP clients can use the hosted endpoint at `https://scrapi.ai/mcp`, or an operator can run a local HTTP instance. The hosted and self-hosted modes use an MCP endpoint rather than exposing a separate scraping API to the client.

## Setup and configuration

Running the scrapi MCP server requires a Scrapi MCP account and API key. The documented setup creates an account through Scrapi’s website and provides an MCP dashboard with an API key. Node.js 20 or newer is needed for the stdio installation path; the hosted Streamable HTTP endpoint does not require Node.js on the client side.

The standard configuration runs `npx -y @scrapi.ai/mcp-server` and passes `SCRAPI_API_KEY` in the environment. The key can also be supplied with the package’s `--api-key` argument. Source installation involves cloning the repository, running `npm install`, and building with `npm run build`, after which the generated Node.js entry point can be configured.

For self-hosted HTTP, the package provides an HTTP command and starts on port 3000 by default, with the MCP route at `/mcp`. `PORT` and `HOST` can change the listener settings. A `/health` endpoint is available for checking the local service.

## Tools and capabilities

The scrapi MCP server documents one tool:

- `scrape_url`: Fetches a webpage and returns cleaned content.
- `url`: Required string containing the page to retrieve.
- `format`: Optional value of `markdown` or `text`; Markdown is the default.

The returned page may include metadata and content formatted for LLM input. JavaScript rendering, clutter removal, retry handling, and anti-bot processing are part of the Scrapi service described by the project. Account usage includes credit tracking and usage analytics through MCP keys, according to the README.

## Limitations and notes

Scraping requires a valid Scrapi MCP account and API key, and usage is tied to the account’s credit allowance. The README describes a Free plan with 500 credits per month, but does not specify how many credits individual requests consume or guarantee that every URL can be retrieved.

The stdio path depends on Node.js, while HTTP avoids that client-side requirement. Claude Desktop needs the `mcp-remote` proxy when connecting to the hosted HTTP endpoint. Client configuration and restart steps vary by application, so the relevant MCP configuration must be added to the selected client before the tool becomes available.

_Full upstream README: https://allmcps.com/mcp/scrapi/readme_