MCP Search Setup for Cursor and VS Code

Adding web search to Cursor, VS Code, or Windsurf via MCP takes one configuration entry and gives your coding assistant access to live documentation, changelogs, and Stack Overflow without leaving the editor. The setup takes under two minutes and costs $0.005 per search query.

Why IDE Agents Need Search

Coding assistants like Cursor and GitHub Copilot have strong code generation capabilities but limited knowledge of recent changes. When a library releases a breaking change, the model's training data is months behind. A web search tool bridges this gap: the agent searches for current documentation before generating code that uses deprecated APIs.

Common failure modes without search: generating code for Next.js 14 APIs when the project uses Next.js 15, suggesting deprecated React patterns, using removed CLI flags, referencing outdated library interfaces. Each of these requires manual correction that a search query would have prevented.

MCP Setup for Cursor

Cursor supports MCP servers natively. Add the search server to your project's .cursor/mcp.json or global configuration:

{
  "mcpServers": {
    "scavio-search": {
      "url": "https://mcp.scavio.dev/mcp",
      "headers": {
        "x-api-key": "your_scavio_api_key"
      }
    }
  }
}

After saving, restart Cursor. The search tool appears in the agent's available tools. When you ask Cursor to look up documentation or check for breaking changes, it will use the search tool automatically.

MCP Setup for VS Code (Copilot Chat)

VS Code's Copilot Chat supports MCP servers through the mcp.json configuration:

{
  "mcpServers": {
    "scavio-search": {
      "url": "https://mcp.scavio.dev/mcp",
      "headers": {
        "x-api-key": "your_scavio_api_key"
      }
    }
  }
}

Place this in your project's .vscode/mcp.json. The search tool becomes available to Copilot Chat's agent mode, which can search the web, read results, and incorporate findings into code suggestions.

MCP Setup for Claude Code CLI

Claude Code reads MCP configuration from .mcp.json in the project root:

{
  "mcpServers": {
    "scavio-search": {
      "url": "https://mcp.scavio.dev/mcp",
      "headers": {
        "x-api-key": "your_scavio_api_key"
      }
    }
  }
}

Verifying the Setup

After configuration, test by asking the agent a question that requires current information:

• "What are the breaking changes in Next.js 15.3?"
• "Show me the current Prisma migrate syntax"
• "What's the latest stable version of Bun?"

If the agent responds with current, accurate information (not training-data answers), the search tool is working. If it responds with outdated information, check that the MCP server URL is correct and the API key is valid.

Cost Management

In a typical coding session, an agent makes 5-20 search queries. At $0.005/query, that is $0.025-$0.10 per session. The 250 free monthly credits cover roughly 10-50 coding sessions, which is enough for most individual developers. Teams with multiple developers sharing an API key should monitor usage via the Scavio dashboard.

Tavily and Exa also offer MCP servers with similar setup. Tavily provides 1,000 free searches/month on their Researcher plan, making it a strong free-tier option. Exa provides 1,000 free searches/month with a semantic search approach that excels at finding conceptually related documentation. The choice depends on whether you prefer structured SERP results (Scavio), summarized answers (Tavily), or semantic discovery (Exa).

Permission Considerations

MCP servers make HTTP calls from your machine. Ensure your API key has appropriate scoping (read-only search access, no write permissions). Store keys in environment variables rather than hardcoding them in mcp.json files that might be committed to version control. For team setups, use separate API keys per developer to track usage and revoke access independently.