Skip to main content

​
Overview

LiteLLM ships a built-in MCP gateway that lets the LiteLLM Proxy connect to Model Context Protocol servers and surface their tools to any model you route through it. ScrapeGraphAI is available as a first-party MCP server, so a single config entry gives every LiteLLM client access to smart scraping, web crawling, search scraping, and agentic scraping workflows. LiteLLM ships ScrapeGraph in its default mcp_servers.json, pointing at the ScrapeGraph MCP server hosted on Smithery.

LiteLLM MCP docs

How LiteLLM connects to MCP servers

ScrapeGraph MCP server

The MCP server LiteLLM connects to

​
Prerequisites

pip install 'litellm[proxy]'

​
Configure the MCP server

Add ScrapeGraph to the mcp_servers block of your LiteLLM proxy config. This is the same entry LiteLLM ships in its default mcp_servers.json — the Smithery-hosted server exposes both HTTP and SSE transports.
config.yaml
model_list:
  - model_name: gpt-5
    litellm_params:
      model: openai/gpt-5
      api_key: os.environ/OPENAI_API_KEY

mcp_servers:
  scrapegraph:
    url: "https://smithery.ai/api/mcp/scrapegraph-mcp"
    description: "Smart scraping, web crawling, search scraping, and agentic scraping workflows."
The raw mcp_servers.json entry added in LiteLLM looks like this:
mcp_servers.json
{
  "scrapegraph": {
    "http_url": "https://smithery.ai/api/mcp/scrapegraph-mcp",
    "sse_url": "https://smithery.ai/api/mcp/scrapegraph-mcp/sse",
    "description": "The ScrapeGraph MCP server provides programmatic access to ScrapeGraph AI's web scraping capabilities, including smart scraping, web crawling, search scraping, and agentic scraping workflows."
  }
}
The ScrapeGraph MCP server reads your ScrapeGraphAI API key from SGAI_API_KEY. Set it in the environment where the proxy runs, or pass it through the Smithery config of your MCP server deployment.

​
Start the proxy

export OPENAI_API_KEY="your-openai-key"
export SGAI_API_KEY="your-scrapegraph-key"

litellm --config config.yaml
The proxy boots on http://localhost:4000 and registers the ScrapeGraph tools under the MCP gateway.

​
List the available tools

LiteLLM exposes connected MCP tools over its MCP endpoint. Point any MCP-aware client at http://localhost:4000/mcp to discover them: 
curl -s http://localhost:4000/mcp \
  -H "Authorization: Bearer $LITELLM_API_KEY"
The ScrapeGraph server registers these tools:
ToolWhat it does
scrapeFetch a page as markdown, HTML, links, or a screenshot
extractExtract structured JSON from a URL with a prompt and optional schema
searchSearch the web and return ranked results
crawl_startStart an async multi-page crawl
crawl_get_statusPoll a crawl job’s progress
crawl_stop / crawl_resumeControl an active crawl
schemaGenerate or augment a JSON Schema from a prompt
monitor_*Create, list, pause, resume, and inspect scheduled jobs
creditsCheck remaining account credits
historyView paginated request history

​
Use it from a model

With the gateway running, any model routed through LiteLLM can call the ScrapeGraph tools during a completion. Pass the proxy’s MCP tools through your client of choice: 
import openai

client = openai.OpenAI(
    base_url="http://localhost:4000",
    api_key="sk-1234",  # your LiteLLM proxy key
)

response = client.responses.create(
    model="gpt-5",
    input="Scrape https://scrapegraphai.com and summarize what the product does.",
    tools=[
        {
            "type": "mcp",
            "server_label": "scrapegraph",
            "server_url": "http://localhost:4000/mcp",
            "require_approval": "never",
        }
    ],
)

print(response.output_text)
The model decides when to call scrape, search, or extract, receives the structured result, and writes its final answer from the scraped data.

​
Support

GitHub Issues

Report bugs and request features

Discord Community

Get help from our community