How to Give Claude Access to Your Computer
Computer access for Claude covers three separate things: controlling the screen, reading local files, and running terminal commands. Each has its own setup and its own risk. Here is how to set each one up and what to decide before you do.
Updated 6 min read
"Computer access" for Claude means at least three different things: controlling what appears on the screen, reading files in a specific folder, and running commands in a terminal. Each is set up differently, carries different risk, and is appropriate for different kinds of work.
Before enabling any of these, decide which kind you actually need and set explicit boundaries around it. Tools like fenbs support this by keeping the approved scope and any standing rules where Claude can read them at the start of each session, so the boundaries are explicit and consistent.
Three kinds of computer access
Understanding the three types helps you grant the minimum access needed and avoid surprises.
| Kind | What it means | How to set it up |
|---|---|---|
| Screen and mouse control | Claude controls the browser or desktop UI, clicking and typing as if it were a person | Available in Claude.ai on higher-tier plans via the computer use feature; the UI shows a live screen Claude is operating |
| Local file access | Claude reads (and possibly writes) files in a specific directory on your computer | Set up in Claude Code by opening a project folder; Claude Code operates within that directory by default |
| Terminal and commands | Claude runs shell commands, scripts, and tools on your machine | Available in Claude Code; Claude proposes and runs commands with your confirmation or pre-approved permissions |
Most people who search for "computer access" are looking for file and terminal access via Claude Code, not screen control. Screen control is a distinct feature, available only on certain plans—check Claude.ai's current plan page for availability.
Giving Claude access to local files with Claude Code
Claude Code runs in a terminal on your machine. When you open Claude Code in a project folder, it reads the files in that directory and can propose edits, read configuration, and run the project's scripts. It does not reach outside that folder without you explicitly asking it to.
To start, install Claude Code through its official setup guide, then open your terminal in the project folder you want Claude to work in. Claude Code reads the directory structure, existing files, and configuration automatically. You can also specify individual files explicitly in your prompt.
Claude Code asks for confirmation before making changes that are not trivially reversible. Keep this confirmation step active, especially early in a project. See what Claude Code is for a full introduction.
Terminal access: what Claude can run
Claude Code can run commands in your terminal—package installs, test runners, build scripts, file operations—either with your approval for each command or with permissions you have pre-approved for common operations like running tests.
Pre-approving broad categories of commands (such as all npm commands) is convenient but reduces the visibility you have over what is happening. Start with per-command confirmation and move to broader permissions only for operations you have seen run correctly several times. Avoid pre-approving commands that delete files, modify system configuration, or send network requests to services you do not recognize.
Scope matters
Claude Code works on what is in the current folder by default. If you open a sensitive directory—your home folder, a folder with credentials—those files are within reach. Work in a dedicated project folder rather than a directory that contains everything.
Screen control: how computer use works
Claude's computer use feature gives Claude access to a browser or desktop interface it can operate by clicking, typing, scrolling, and navigating menus. This is significantly more powerful than file access—Claude can interact with any application visible on screen, including web apps, desktop software, and settings panels.
Computer use is designed for tasks that require navigating a UI rather than writing code: filling in a form, checking a result in a browser, or working with a tool that has no API. It is less suited to tasks where code or direct API calls would be faster and more reliable.
Never start a computer use session with sensitive accounts signed in unless Claude specifically needs access to them. Check which accounts are active in the browser before starting, and log out of anything unnecessary. Watch the session live rather than leaving it to run unattended.
Record approved work scope before Claude starts
fenbs lets you write the approved scope—which directories, which commands, which accounts—as Decisions and rules before each session starts. Claude reads these rules at the start of the session via MCP, so the boundaries are explicit and consistent without you having to state them every time. Connect Claude Code from Integrations and add the scope decisions to your project.
Create a free accountFree for one person · No credit card · Works with Claude, ChatGPT, Cursor and more
What to decide before granting access
The single most useful question is: what is the minimum access Claude needs to do this specific task? File access to one project folder is safer than access to the home directory. Approval for test-running commands is different from approval for deployment commands.
- Which directory is in scope? Open Claude Code inside that folder, not above it.
- Which commands need pre-approval versus per-command confirmation? Default to confirmation.
- Which accounts should be logged out before a computer use session? Log them out before starting.
- Who approves changes before they are applied to a shared or production resource? A person, not Claude.
- What is the reversibility of the actions Claude might take? Prefer reversible actions first.
How to do this with fenbs
- Create a free account and add the project to your fenbs board. Give it a description that includes which system it operates on.
- Write the scope decisions in Decisions and rules: which directory, which commands are approved, what requires human sign-off before Claude proceeds.
- Connect Claude Code via MCP from Integrations. At the start of each session, Claude reads the approved scope before it starts working.
- Open Claude Code in the specific project folder—not your home directory or any folder with credentials outside the project.
- Review Claude's proposed commands before approving them, especially the first time you try a new kind of task. Save any new scope decisions in fenbs so the next session starts with an accurate picture of what is approved.