SuperAPI AI Directory
Partner with us中文AI newsFavorites
All toolsAI developer platformsdecodo-mcp-server

decodo-mcp-server

Easy web data access. Simplified retrieval of information from websites and online sources.

Open website

About this tool

Purpose

Decodo MCP Server is a web scraping layer for AI agents. Through MCP (Model Context Protocol), it connects MCP-compatible clients to Decodo's Web Scraping API. Its main uses include: providing web scraping capabilities for LLMs, providing real-time data retrieval for RAG, supporting AI agent browsing and research, and extracting structured data from dynamic websites. Users do not need to maintain their own proxies, parsers, and retry logic; a single integration point is enough to obtain reliable web data access.

Main Capabilities

  • Web scraping for AI agents, with no infrastructure required: can scrape any website (including JavaScript-heavy pages) without having to handle proxy rotation, CAPTCHA solving, or anti-bot systems yourself.
  • Structured output for LLM workflows: Markdown (suitable for LLMs), JSON (suitable for structured pipelines), and screenshots (providing visual context), which can be used in RAG pipelines, AI research agents, and automated processes.
  • Built-in support for popular targets: ready-made tools for Google and Bing (search engine results pages), Amazon, Walmart, and Target (e-commerce), Reddit, TikTok, and YouTube (social media), as well as ChatGPT and Perplexity (AI search).
  • Global proxy infrastructure: more than 125 million residential IPs, covering over 195 geographic locations, achieving a 99.99% success rate even for heavily protected targets.
  • Modular MCP toolset: web, search, ecommerce, social_media, and ai can be enabled as needed, making tool selection clearer and agent performance better.
  • Quick start: from obtaining an API key to completing the first scrape takes only a few minutes, with no additional setup required.

How to Use

Prerequisites: create a free account at dashboard.decodo.com (up to 2K free requests, no credit card required); obtain an API key (copy the Web Data API key from the Web Data API subscription in the dashboard; the basic authentication token provided by older plans also works); install Node.js 18+; prepare an MCP client (such as Claude Desktop, Cursor, Windsurf, or other MCP-compatible tools); configure the MCP server in an AI client.

Configuration example (the local server reads credentials from SCRAPER_API_TOKEN and automatically detects the type):

  • Claude Desktop: Open Settings → Developer → Edit Config, add the configuration to claude_desktop_config.json, save, and restart.
  • Cursor: Open Settings → MCP, click Add a new global MCP server (which opens mcp.json), add the same configuration, then save and check for the green status indicator next to Decodo.
  • Windsurf: Open Settings → Windsurf Settings, scroll to Cascade → Add custom server + (which opens mcp_config.json), save, and restart Windsurf.

Once connected, you can test it: enter the prompt "Scrape the titles of the top 5 articles from Hacker News" in the client, and you should get a structured list within seconds. If an authentication error occurs, check the API key in the dashboard.

Running this MCP server locally: an MCP client is required (commonly Claude Desktop and Cursor). The steps are: clone the repository; run the command in the terminal; note the build location (append index.js to the end of that directory; the build file location will look like the corresponding path); update the MCP client with the server information.

Enable specific tool sets as needed: tools are organized by tool set, and a comma-separated list can be passed via the toolsets query parameter to selectively enable them. If no tool set is specified, all tools are registered.

Limit the number of response tokens: if the agent's context window is small, scraped content will be automatically truncated to avoid context overflow; you can increase the number of returned tokens in the prompt. If the agent's context window is large, you can request the full content.

Scraping geo-restricted content: by giving the AI agent the corresponding prompt, it will learn that a certain site (for example, peacock.com) has geographic restrictions, and the geographic restrictions can be bypassed in a corresponding way.

Using with Decodo agent skills: this serverprovides scraping tools for agents, while Decodo/agent-skills teaches agents when to use them, which interface to use, and how to invoke them, so they autonomously choose the correct tools and parameters. The skills can route between the decodo CLI, the hosted MCP server, and the raw HTTP API, and automatically fall back to the MCP server when no shell is available. Together, they provide both the execution layer (this server) and the decision layer (the skills). For a Claude Code quick start, see the repository instructions; for manual installation and other agents (Cursor, Codex, Gemini CLI, Windsurf), see the agent-skills README.

Use cases

  • AI-driven web scraping: let LLMs collect the latest data instead of relying on static training data.
  • RAG with real-time data: pull real-time Google, Bing, and AI search results into retrieval pipelines.
  • E-commerce intelligence: track product prices, listings, and sellers across markets without being blocked.
  • Social media data collection: collect posts, channels, and engagement data from Reddit, TikTok, and YouTube.
  • Travel and price aggregation: build tools that collect real-time prices and availability across websites.

Limitations

The material does not provide specific pricing plans, request rate limits, a complete list of supported languages, or specific parameter details for each tool; scraping results may be affected by the target website's anti-scraping strategies and geographic restrictions, requiring corresponding proxies or bypass methods. The code is released under the MIT License.

Related tools

Browse category