How to Use ChatGPT to Work With Files on Your PC: A Complete Guide

Conclusion first

AI applications such as ChatGPT can read, edit and work with files on your computer when the relevant features and permissions are available. For example, they may summarize documents, organize folders or help edit and test code.

What an AI can actually do depends on its capabilities, execution environment, connection state and permissions. Uploading a file to a chat is different from letting an AI work inside an authorized local environment or send a tool request to your own PC.

This guide compares those approaches, focusing on where work runs, which files are accessible and who controls the permissions.

Overview

There are three common ways to let AI work with PC files.

Method 1: Give the files to the AI

You can upload a spreadsheet or PDF to a chat and ask for analysis, then download the output. The AI normally works with the information you supplied or explicitly connected. It does not automatically gain permission to browse unrelated folders or edit your original local files.

Method 2: Use an AI working environment

Some AI tools can operate in an authorized local folder or development environment. On supported desktop configurations, ChatGPT Work can work with permitted local files and applications. Codex can edit code and run tests in local or cloud development environments.

The distinction between local and cloud execution matters because it changes which files, tools and permissions are available.

Method 3: Connect the AI to your PC through external tools

A connection service can receive requests from the AI, execute allowed operations on the user’s PC and return the results. For example, an AI can request a folder listing and then summarize it.

OGINAU Remote is being developed as a connection method for using a user’s own Windows PC as an authorized execution destination.

Key topics

What work becomes possible?

Consider a monthly reporting workflow. An employee might open a folder, review files, extract figures, write a report and save the result. With the necessary authorized tools, an AI could assist with several steps:

1. Find the relevant files in an approved folder
2. Read the authorized data
3. Organize the information into a report
4. Save the output to a permitted destination

This is an illustrative workflow, not a claim that every AI product supports every step automatically.

Five components of an MCP-to-Windows connection

1. MCP defines the operations available to AI. Model Context Protocol is a common protocol connecting AI applications with external tools and data. A provider can expose operations such as listing folders, reading files and writing files. MCP does not itself grant unrestricted access to a PC. See What is MCP?.

2. Authentication and authorization verify identity and permissions. A valid login does not imply permission to edit every file or use somebody else’s computer.

3. A Gateway routes authorized requests to the right computer. It helps identify the user and target device and check the permitted operation.

4. A Windows Connector performs the requested operation on the PC. A local component executes permitted file operations and returns the result.

5. Allowed Root limits the accessible area. The connection can be restricted to an approved work folder. File moves should respect both source and destination boundaries, while writes should account for concurrent changes.

How a request travels

Suppose someone asks an AI: “List the files in my work folder.”

1. The user requests the task.
2. The AI chooses a relevant available tool.
3. The request travels through an MCP connection.
4. The Gateway checks the user, device and permissions.
5. The Windows Connector lists files within the allowed folder.
6. The result returns to the AI for a readable answer.

If the computer is offline or the required permission is missing, execution should fail safely. The goal is not unlimited access but controlled execution of explicitly permitted operations.

MCP, Computer Use and remote desktop

Method	How it works	Typical tasks
MCP tool connection	Calls structured, predefined operations	File handling, search, APIs
Computer Use	AI interprets and interacts with a graphical interface	GUI-focused tasks
Remote desktop	Displays and controls a remote computer screen	Human-driven remote operation

These approaches can complement one another. The best choice depends on the job and the execution boundaries.

What implementation and testing showed

OGINAU is developing OGINAU Remote to connect AI applications to a user’s own Windows PC. Its internal engineering and test records document:

• End-to-end Windows connection: five initial tests passed through a Gateway and Windows Connector.

• MCP file-operation path: an HTTP MCP request reached the Windows-side execution path through the Connector.

• Folder restrictions: Allowed Root checks were implemented and tested.

• Concurrent update protection: a SHA-256-based guard rejected a write based on a stale file state.

• Permission separation: public-facing file tools were distinguished from more powerful internal process capabilities.

These are results from development tests, not a guarantee of current general availability or operation in every environment.

The objective is to connect existing AI tools to an authorized Windows execution environment, not to build a new AI model.

Safety considerations

Before enabling AI access to a PC, consider the following:

• Identity: Can the service identify who requested an operation?

• Permissions: Can reading and writing be controlled separately?

• Folder boundaries: Can it prevent access outside approved folders?

• Revocation: Can an unnecessary device connection be disabled?

• Auditability: Can requests and their outcomes be reviewed?

• Mistake prevention: Are important changes protected?

Using MCP does not automatically make an integration secure. Authentication, device ownership, operation permissions and filesystem boundaries have to work together.

Sources and primary information

This article distinguishes public standards from OGINAU’s internal development and verification records.

• Model Context Protocol specification: MCP architecture, Tools, Client and Server roles.

• OpenAI: ChatGPT Work and Codex: AI work environments and different execution capabilities.

• OGINAU Remote engineering records: Windows Connector and Gateway tests, MCP file operations, Allowed Root controls and SHA-256 stale-write rejection.

Private credentials and user-specific information are not included.

What to read next

• What is MCP? A beginner’s guide

Understanding MCP helps explain how AI applications connect not only to PCs but also to other services.

Summary

AI applications such as ChatGPT can work with PC files through file uploads, supported local working environments or authorized external tool connections.

The key questions are where execution takes place, who controls the environment and which operations are allowed.

OGINAU is developing and testing a way to connect a user’s Windows PC as an authorized execution environment for existing AI tools.