Give Claude Code, Cursor or Claude Desktop Live Web Access with MCP

Tool list and the scrape tool were exercised over Streamable HTTP on 4 November 2026.

Language models know a lot about the web as it was when they were trained. They know nothing about the page you just published or the price a competitor changed this morning. The Model Context Protocol (MCP) fixes that by letting an assistant call tools you provide, and a web-data tool server is one of the most useful ones to add.

#What MCP is, in one paragraph

MCP is an open protocol for connecting AI applications to tools and data. A client (Claude Code, Cursor, Claude Desktop and many others) connects to a server, asks which tools it offers, and lets the model call them during a conversation. Our server speaks MCP over Streamable HTTP, so there is nothing to install on your machine for clients that support remote servers.

#Setup

You need an API key from the dashboard (API keys → Create key). Every client below uses the same endpoint:

https://api.lucidextractor.com/mcp-http/mcp

The server accepts the key as Authorization: Bearer <key> or as an X-API-Key header.

Claude Code

claude mcp add --transport http lucidextractor \
  https://api.lucidextractor.com/mcp-http/mcp \
  --header "Authorization: Bearer $LUCID_API_KEY"

Run claude mcp list to confirm it connected, then ask something like "Read https://docs.python.org/3/tutorial/classes.html and summarise how instance variables work."

Cursor

Add this to .cursor/mcp.json in your project (or ~/.cursor/mcp.json for every project):

{
  "mcpServers": {
    "lucidextractor": {
      "url": "https://api.lucidextractor.com/mcp-http/mcp",
      "headers": { "Authorization": "Bearer YOUR_API_KEY" }
    }
  }
}

Keep the file out of version control if it contains the key.

Claude Desktop

Claude Desktop's config file launches local servers, so use the small mcp-remote bridge to reach a remote one. Edit claude_desktop_config.json (on macOS: ~/Library/Application Support/Claude/; on Windows: %APPDATA%\Claude\):

{
  "mcpServers": {
    "lucidextractor": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://api.lucidextractor.com/mcp-http/mcp",
        "--header",
        "Authorization:${LUCID_AUTH}"
      ],
      "env": { "LUCID_AUTH": "Bearer YOUR_API_KEY" }
    }
  }
}

The header value goes through an environment variable because some systems split arguments on spaces. Restart Claude Desktop after saving.

#The 9 tools, grouped by job

#What a tool call returns

scrape returns the page as markdown by default (output_format also accepts html, text and json). Set render_js: true for pages that only fill in after JavaScript runs. If a page cannot be fetched, the tool returns success: false with the reason. That gives the model something true to say instead of summarising an error page as if it were the content.

#Prompts that work well

• "Map https://fastapi.tiangolo.com and list the tutorial pages about request bodies, then read the two most relevant ones."
• "Crawl https://example.com up to 25 pages and list the pages that have no meta description."
• "Run an SEO analysis of https://example.com and tell me the three fixes with the biggest impact."

#Treat page content as untrusted

Anything a tool returns comes from a web page, and web pages can contain text written to manipulate an AI ("ignore your previous instructions and..."). A few habits keep you safe:

• Keep tools that can act on your behalf (sending email, writing files, running commands) behind confirmation when web content is in the conversation.
• Prefer read-only tools for research tasks.

#What it costs

Each tool call spends the same credits as the matching API request. MCP calls draw on the same credit balance and the same plan rate limits as the REST API.