Welcome back to Advanced Claude Code Features! In the previous units, we explored custom slash commands, managed tool permissions, and learned to control workflows with Plan Mode and Task delegation. Now, we will expand beyond the local codebase to access current information from the web and optimize how tools are used.

In this lesson, we will explore Web Access and Tool Optimization. Web tools enable us to search for current information and retrieve specific documentation pages, while understanding tool selection and batching helps us write more efficient workflows. These capabilities are particularly valuable when working with rapidly evolving frameworks or when we need to reference external resources.

By the end of this lesson, we will understand when to use WebSearch versus WebFetch, how to configure web access permissions, why Claude uses specialized tools instead of generic shell commands, and how tool batching improves performance.

When we work with modern frameworks and libraries, documentation changes frequently. What we learned six months ago might be outdated, and critical features from the latest release might not be in Claude's training data. This is where web tools become essential.

Claude Code provides two complementary web tools. WebSearch queries the web and returns summaries of results, similar to using a search engine. It's designed for discovery: finding the right resource, checking current best practices, or getting quick facts. WebFetch retrieves the full content of a specific page, similar to opening a URL in your browser. It's designed for deep reading: analyzing documentation, understanding implementation details, or reviewing complete blog posts.

These tools work together naturally: we often use WebSearch to discover resources, then WebFetch to read them in detail. Let's see how each one works in practice.

When we need current information or want to find relevant resources, WebSearch provides targeted results. The tool takes a query string and returns summaries from multiple sources:

Notice how WebSearch provides a concise summary. We don't get full page content, but we get enough information to:

• Verify current recommendations
• Discover new features or tools
• Find relevant documentation pages

The search results help us understand the landscape before diving deeper. When we find something worth exploring in detail, that's when WebFetch becomes valuable.

Once we've identified a specific resource, WebFetch retrieves the complete page content. This is particularly useful for reading documentation, analyzing blog posts, or understanding implementation details:

WebFetch extracts the main content from the page, removing navigation, ads, and other non-essential elements. This gives Claude the full context needed to:

• Answer detailed questions about the content
• Compare multiple documentation sources
• Extract code examples or API signatures

The tool returns substantially more information than WebSearch, which is why we use it strategically for specific pages rather than broad exploration.

For security and control, we can configure which domains WebFetch can access. This is particularly important in team environments or when working with sensitive codebases. The configuration lives in .claude/settings.json:

These rules create a domain allowlist:

• domain:github.com permits fetching from GitHub repositories.
• domain:*.react.dev permits any subdomain of react.dev (docs.react.dev, beta.react.dev, etc.).
• Any other domain would require explicit approval.

This configuration helps maintain security while enabling access to trusted documentation sources. When Claude attempts to fetch from a non-allowed domain, it will ask for permission rather than proceeding automatically.

Understanding when to use each tool improves our workflow efficiency. The decision depends on what information we need:

These tools often work together in sequence:

1. Start with WebSearch to discover relevant resources.
2. Use WebFetch to read promising pages in detail.
3. Iterate based on what we learn.

For example, searching for "React 19 migration guide" might reveal several resources; then, we fetch the official migration guide for complete implementation steps.

Beyond web tools, Claude's approach to tool selection follows an important principle: Use specialized tools instead of generic shell commands. This isn't arbitrary; specialized tools are more reliable, safer, and better integrated with Claude's workflow.

Consider file operations. We might be tempted to use Bash commands like cat, sed, or find because they're familiar. However, Claude's native tools offer significant advantages:

The specialized tools handle edge cases automatically: special characters, encoding, atomic operations, and structured output. This reduces errors and makes workflows more predictable.

Let's examine why native tools are preferred for reading files. When Claude uses the Read tool, it handles encoding automatically and returns structured results:

Compare this to using Bash:

The Read tool provides:

• Automatic encoding detection and handling.
• Structured metadata (line counts, file info).
• Consistent output format.
• Better error handling.

Similarly, Write creates files atomically (reducing corruption risk), and Edit applies changes with validation. These guarantees matter when making automated changes to important files.

The same principle applies to searching: Grep and Glob are specialized tools that outperform their Bash equivalents:

Grep returns structured results with file paths, line numbers, and context. Glob efficiently handles pattern matching across directory trees. Both integrate smoothly with other tools and avoid shell escaping pitfalls.

The guideline is clear: use Bash only for operations that truly require it, such as:

• Package management: npm install, pip install
• Version control: git commit, git push
• Running build tools: npm run build, pytest
• Process management: checking running processes

For file operations, search, and pattern matching, native tools are the correct choice.

When operations are independent (don't depend on each other's results), Claude can batch them together. This significantly reduces latency by minimizing API roundtrips:

Notice the execution pattern: three Read operations were requested simultaneously and executed in parallel. The results arrived together, reducing total time from what would have been three sequential roundtrips to a single batched operation.

Batching benefits include:

• Fewer API roundtrips: Multiple operations complete in one request/response cycle.
• Faster execution: Parallel processing where possible.
• Lower latency: Especially important over network connections.

Claude automatically batches operations when it determines they're independent. We don't need to do anything special; the optimization happens transparently when multiple tool calls don't have dependencies between them.

An important detail about the Bash tool: it maintains a persistent working directory across multiple calls, but environment variables and other shell state do not persist:

The shell session maintains:

• Directory changes: cd commands affect subsequent calls.
• Working directory state: The current directory persists.

But does NOT maintain:

• Environment variables set with export
• Aliases or functions
• Other shell state

For operations requiring environment variables or complex shell state, combine commands in a single call:

• export VAR="hello" && echo $VAR - works in one call
• npm install && npm test - runs tests after installation
• git add . && git commit -m "..." - stages and commits in sequence
• docker build && docker run - builds, then runs containers

Understanding directory persistence helps us write multi-step workflows where navigation matters, while command chaining lets us handle operations that need shared shell state within a single execution context.

In this lesson, we explored web access capabilities and tool optimization strategies. WebSearch helps us discover current information through targeted queries, while WebFetch retrieves complete pages for detailed analysis. We learned to configure domain restrictions for security, and we understood when each tool is appropriate.

We also examined why Claude prefers specialized tools over generic Bash commands: native tools like Read, Write, Edit, Grep, and Glob offer better reliability, structure, and integration. Tool batching automatically optimizes independent operations for lower latency, and Bash session persistence enables multi-step shell workflows.

These optimization principles help us work more efficiently: web tools keep our knowledge current, specialized tools reduce errors, and batching minimizes roundtrips. Now it's time to practice! The upcoming exercises will challenge you to apply these concepts in real scenarios, choosing the right tools and understanding the trade-offs. Get ready to build more efficient Claude Code workflows!