Claude Code has a set of built-in tools it uses to interact with your workspace and the wider internet. These tools are what allow Claude to go beyond answering questions in text -- they let it open files, make changes, search your workspace, run system commands, and look things up online. You don't call these tools directly. Claude chooses the right tool based on what you ask it to do and does the work in the background. Understanding them helps you know what Claude can and cannot do, and how to ask for things effectively.
File tools
These are the tools Claude uses most frequently. When you're working with documents, reports, spreadsheets, or any other files in your workspace, these four tools are doing the work.
Read -- opens files
The Read tool opens a file and loads its contents so Claude can work with it. Claude uses Read whenever you ask it to look at something that already exists.
You say: "Look at my invoice template and tell me if anything is missing."
Claude does: Uses Read to open the invoice file, then reviews its contents and responds.
Read is non-destructive -- it cannot change anything. It's always safe. Claude can read text files, code files, Markdown documents, and many other formats. It can also read specific sections of large files rather than loading the whole thing, which keeps sessions running efficiently.
Write -- creates new files
The Write tool creates a brand-new file, or completely replaces an existing file with new content. Claude uses Write when you ask it to create something from scratch.
You say: "Create a meeting notes template I can reuse each week."
Claude does: Uses Write to create a new file with the template content at the location you specify.
Because Write can overwrite existing files, Claude will typically confirm what it's about to write before doing so -- unless you've pre-approved the action in your settings.
Edit -- modifies existing files
The Edit tool makes targeted changes to a file that already exists -- replacing specific text rather than rewriting the whole thing. Claude prefers Edit over Write when modifying existing content, because it's more precise and leaves everything else untouched.
You say: "Change the company name in that document from Acme Ltd to Acme Group."
Claude does: Uses Edit to find the old name and replace it, leaving everything else in the document unchanged.
Glob -- finds files by name
The Glob tool searches your workspace for files matching a pattern. "Glob" is a technical term for a pattern that uses wildcards to match multiple filenames -- for example, "all files ending in .pdf" or "all files in the invoices folder."
You say: "Find all the PDF invoices from last year."
Claude does: Uses Glob to search for files matching the pattern invoices/2025/*.pdf and returns the list.
Glob returns file paths -- it doesn't open the files. Claude typically uses it first to locate the right files, then uses Read to open them.
Grep -- searches inside files
The Grep tool searches the contents of your files for specific words or phrases. Where Glob finds files by name, Grep finds files by what's inside them.
You say: "Which file mentions the Johnson contract?"
Claude does: Uses Grep to search the contents of files in your workspace for "Johnson contract" and returns which files contain a match, with the relevant lines shown.
Grep is faster and more precise than asking Claude to read every file one by one. It's especially useful when you have many files and aren't sure where something is recorded.
Execution tools
These tools let Claude do things on your computer beyond reading and writing files.
Bash -- runs terminal commands
The Bash tool runs commands in your computer's terminal (the command-line interface -- the text-based window where you can run programs and scripts). This is what gives Claude the ability to do things like save your work to version control, install software, run automated scripts, or check system status.
You say: "Commit today's work to GitHub and push it."
Claude does: Uses Bash to run git add, git commit, and git push commands in sequence.
Bash is powerful and Claude will typically show you what command it plans to run before executing it. You can pre-approve safe commands (like git status or running tests) so Claude doesn't need to ask each time, while keeping dangerous commands (like deleting files) behind a confirmation step.
Agent -- spawns a parallel worker
The Agent tool creates a separate Claude instance -- called a subagent -- that works on a specific task independently, in its own context window. A context window is the amount of information Claude can hold in its working memory at one time. Spawning a subagent means the main conversation stays lean while heavy work happens elsewhere.
You say: "Research these five competitors and summarise what each one offers."
Claude does: Uses Agent to spawn five parallel subagents, one per competitor. Each researches independently and writes its findings to disk. Claude reads the summaries and combines them for you.
You'll rarely need to ask Claude to use agents explicitly -- it decides when the task benefits from parallel or delegated work. But knowing this tool exists helps explain why Claude sometimes says it's "spawning agents" for a large task.
Web tools
These tools let Claude reach beyond your local files and access information from the internet.
WebFetch -- reads a web page
The WebFetch tool retrieves the content of a specific web page. Claude uses it when you give it a URL and ask it to do something with what's there.
You say: "Check what's on our homepage and tell me if it accurately describes our services."
Claude does: Uses WebFetch to load your homepage URL, reads the content, and compares it against what you've told it your services are.
WebSearch -- searches the internet
The WebSearch tool runs a live search query and returns results. Claude uses it when you ask for current information it wouldn't have in its training data -- prices, news, recent events, or anything that changes over time.
You say: "Find the current pricing for Slack Business plans."
Claude does: Uses WebSearch to query for current Slack pricing and returns the result with the source.
Which tool does Claude use when?
| When you ask Claude to... | Tool used |
|---|---|
| Look at a file, read a document, review a template | Read |
| Create a new file, write a new document from scratch | Write |
| Change something in an existing file | Edit |
| Find a file by name or extension | Glob |
| Find which file contains a specific word or phrase | Grep |
| Save work to version control, run a script, install something | Bash |
| Handle a large, parallel, or time-consuming research task | Agent |
| Check what's on a specific URL or web page | WebFetch |
| Find current prices, recent news, up-to-date information | WebSearch |
What built-in tools cannot do
Understanding the limits is as important as understanding the capabilities. Built-in tools cannot:
- Access databases directly. Claude cannot connect to your CRM, accounting software, or any database unless you've set up a specific integration (called an MCP server) that bridges the connection.
- Interact with graphical interfaces. Claude cannot click buttons, fill in web forms, or operate software that requires a mouse. It works with files and terminal commands -- not with what appears on screen.
- Monitor things in real time. Claude cannot watch a folder for new files, wait for an email to arrive, or check a system status continuously. Each session is a discrete interaction -- Claude acts when you ask it to, not on its own schedule.
- Remember across sessions without a memory system. Without a configured memory tool or well-maintained state files, Claude starts each new session without knowledge of what happened in previous ones.
Extending Claude's capabilities with MCP
The built-in tools cover local files, terminal commands, and the web. For everything else -- your calendar, your email, your accounting software, your CRM -- you connect external tools via MCP (Model Context Protocol) servers. These appear alongside the built-in tools during a session and work the same way: Claude calls them by name, you see what it's doing, and you approve or pre-approve actions. See the MCP guide for setup instructions.
