The MCPs That Made Claude Code Usable for Me

As more and more people start vibe-coding, it becomes even more necessary that their agents are informed of the context they need to perform the requested actions.

I never thought I’d use MCPs until I switched from Cursor to Claude Code. The reason I didn’t need them on Cursor was because of its codebase and documentation indexing. Since other AI agents like Claude Code, Gemini, Codex, etc. don’t have these features built-in, we need MCPs to do this ourselves.

I’ll be sharing the MCPs that I personally find useful and think are a must-have in your repositories.

Disclaimer: I’m not sponsored, no one has paid me, I don’t work for any of these companies.

1. Exa Code MCP

When your AI agent discovers a package it needs more context about, it needs somewhere to refer to. Exa provides search for almost any package or repository you can think of and performs very well.

Install Exa MCP via HTTP method:

{
  "mcpServers": {
    "exa": {
      "type": "http",
      "url": "https://mcp.exa.ai/mcp",
      "headers": {}
    }
  }
}

2. Grep Code Search MCP

Grep is a code search provider for public GitHub repositories. Exa is a generalist, but when you’re using less popular repositories or packages, Grep can help fill in the gaps.

{
  "mcpServers": {
    "grep": {
      "type": "http",
      "url": "https://mcp.grep.app",
      "headers": {}
    }
  }
}

3. Context7 MCP

Context7 provides documentation search for packages and repositories. This is especially useful when your agent needs to refer to official documentation to understand how to use a certain part of a codebase.

Although not required, it’s better to get a free API key from Context7 and add it to the headers for better rate limits.

{
  "mcpServers": {
    "context7": {
      "type": "http",
      "url": "https://mcp.context7.ai/mcp",
      "headers": {
        "CONTEXT7_API_KEY": "API_KEY_HERE"
      }
    }
  }
}

4. DeepWiki MCP

DeepWiki is Cognition Labs’ knowledge base MCP. Devin has indexed most known projects and created AI-parsable wikis that are easy for AI agents to understand when working with open-source packages and projects.

{
  "mcpServers": {
    "deepwiki": {
      "type": "http",
      "url": "https://mcp.deepwiki.com/mcp",
      "headers": {}
    }
  }
}

You must be asking, “why are they all about context?” The reason is simple: the more we depend on AI agents, the more you have to realize that they will hallucinate. LLMs are nothing but a large dataset of random knowledge. To steer them in the right direction, you need to provide specially curated context specific to your prompts. Claude Code was unusable for me without these MCPs. It’d make up implementations before I added them.

I hope this helps someone in their vibe-coding journey! If you have any other MCPs that you think are a must-have, feel free to share them with me on X (Twitter) or via email at [email protected].