Can Claude Code Access the Internet? What It Can and Can’t Do

Claude Code runs in your terminal. It reads your files, writes code, and executes shell commands. But when it comes to the internet, the answer is “sort of.” It has limited built-in web access, and MCP tools fill in the gaps.

What Claude Code Can Do Natively

Claude Code includes a built-in tool called WebFetch. Give it a URL, and it will grab the content from that page. This means if you know the exact URL of a documentation page, a GitHub issue, or a blog post, you can tell Claude Code to fetch it and read the contents.

> Fetch https://docs.python.org/3/library/asyncio.html and summarize the key changes.

This works. Claude Code hits the URL, pulls the HTML, and works with the content. It’s useful when you have a specific link and want Claude Code to read it without you having to copy-paste.

That’s about where the native internet access ends.

What Claude Code Cannot Do

Web search. There is no built-in Google search, Bing search, or any other search engine integration. If you ask “what’s the latest version of React?”, Claude Code answers from its training data, which has a cutoff date. It cannot search the web to check.

Browsing. Claude Code doesn’t have a browser. It can’t navigate between pages, click links, or follow redirects in the way a human would. WebFetch grabs a single page’s content.

JavaScript-rendered content. Many modern websites load their content with JavaScript after the initial page load. WebFetch pulls the raw HTML, so content that requires JS execution won’t appear in what Claude Code sees.

Authentication-gated content. Claude Code can’t log into websites. Anything behind a login wall is off-limits.

The practical result: Claude Code knows what’s in your codebase and what’s in its training data. For anything that requires current information from the web, it’s working blind.

How MCP Tools Change This

The Model Context Protocol is built into Claude Code. You add an MCP server to your configuration file, and Claude Code discovers the tools that server provides. When a tool is relevant to your request, Claude Code calls it automatically.

For internet access, this means you can add tools that give Claude Code:

  • Web search (Google, Bing): search for current information, documentation, error messages, package versions
  • News search: find recent articles and announcements
  • YouTube transcripts: pull the text content from video tutorials and talks
  • Company and product lookups: get structured data about businesses, domains, and products

Once these tools are connected, Claude Code’s behavior changes. Instead of answering “what’s the latest version of X?” from stale training data, it searches Google, finds the answer, and responds with current information.

Setting Up Internet Access with AgentPatch

AgentPatch provides web search and dozens of other internet-connected tools through a single MCP server. One config block gives Claude Code access to Google Search, Google News, YouTube transcripts, and more.

The AgentPatch CLI is designed for AI agents to use via shell access. Install it, and your agent can discover and invoke any tool on the marketplace.

Install (zero dependencies, Python 3.10+):

pip install agentpatch

Set your API key:

export AGENTPATCH_API_KEY=your_api_key

Example commands your agent will use:

ap search "web search"
ap run google-search --input '{"query": "test"}'

Get your API key from the AgentPatch dashboard.

Install the AgentPatch skill — it teaches Claude Code when to use AgentPatch and how to use the CLI:

/plugin marketplace add fullthom/agentpatch-claude-skill
/plugin install agentpatch@agentpatch

MCP Server (Alternative)

If you prefer raw MCP tool access instead of the skill:

claude mcp add -s user --transport http agentpatch https://agentpatch.ai/mcp \
  --header "Authorization: Bearer YOUR_API_KEY"

Replace YOUR_API_KEY with your actual key from the AgentPatch dashboard.

What This Looks Like in Practice

With AgentPatch connected, Claude Code can answer questions it couldn’t before:

> Is there a known issue with Prisma 6.x and Node 22?

Searching Google for "prisma 6 node 22 compatibility issue 2026"...

Yes. Prisma 6.3.0 introduced a regression with Node 22.x on Linux where
connection pooling would stall under load. It was fixed in 6.3.1, released
March 8. If you're on 6.3.0, upgrade to 6.3.1 or later.

Claude Code searched the web, found a current answer, and gave you actionable information. Without the MCP connection, it would have guessed from training data or told you it didn’t know.

You can also combine tools. Ask Claude Code to search for a topic, fetch the top result, and summarize it. The tools compose naturally because Claude Code decides when to call each one based on what you’re asking.

Wrapping Up

Claude Code has limited native internet access: it can fetch specific URLs but can’t search the web. MCP tools remove that limitation. AgentPatch gives you web search and other internet-connected tools in a single config block, no extra API keys needed. More at agentpatch.ai.

← Back to blog