WorkForge
FreeNot checkedWorkForge is a secure Windows MCP gateway that lets ChatGPT inspect local projects, resume Git work, read text and images, apply SHA-guarded edits, and run supe
About
WorkForge is a secure Windows MCP gateway that lets ChatGPT inspect local projects, resume Git work, read text and images, apply SHA-guarded edits, and run supervised PowerShell jobs on the user's workstation through OpenAI Secure MCP Tunnel. It exposes twelve bounded tools while keeping profile identity, runtime evidence, credentials, process ownership, and recovery behavior explicit and locally verifiable.
README
English | 한국어
ChatGPT is smart, but it does not normally have hands inside your PC. WorkForge gives it a safe pair of hands.
WorkForge connects ChatGPT to your Windows workstation so it can inspect real project files, search and make guarded edits, read local images, and run supervised PowerShell commands. If Git is installed, WorkForge can additionally inspect branches, commits, and change history.
In plain language, it is a secure working bridge between ChatGPT and your computer.
With WorkForge connected, you can say things like:
"Read this project and tell me where I left off."
"Find out why this build is failing and fix it."
"Update the README so it matches the current code."
"Check Git and summarize my recent work."
"Run the tests and investigate anything that fails."
Instead of only explaining what you should do, ChatGPT can look at the actual workspace and work through the task with you.
Why does WorkForge exist?
Normally, ChatGPT cannot see the files on your computer.
If a project has a bug, the usual workflow looks something like this:
1. Copy the error message
2. Paste it into ChatGPT
3. Find the related source file
4. Copy that code too
5. Ask for a fix
6. Paste the fix back into the file
7. Run the build yourself
8. Copy the next error
9. Repeat
WorkForge changes that loop to:
You
↓
"Check the project, find the problem, and fix it."
↓
ChatGPT
↓
WorkForge
↓
Your real files · search · optional Git context · PowerShell
ChatGPT can inspect the information it needs, understand the current state, make allowed changes, and verify the result.
That means moving from describing your workspace to AI through copy and paste to letting AI inspect the workspace and collaborate inside it.
What gets better?
1. You spend less time explaining the project
You no longer need to repeatedly describe the folder structure, paste file contents, and report Git state by hand.
Before
"There is a file under Assets/Scripts..."
"Here is the code..."
"I changed this yesterday..."
With WorkForge
"Read the project."
2. ChatGPT can answer from the current state
It can inspect the real files and Git status instead of relying only on old conversation context.
Questions like these become much more useful:
"What was I working on recently?"
"How far is this feature implemented?"
"Why is the build broken right now?"
"Change this file and run the tests afterward."
3. It can move from inspection to verification
WorkForge can do more than read files. It can also run supervised PowerShell jobs.
That enables a practical loop such as:
Inspect code
↓
Edit file
↓
Run build
↓
Read errors
↓
Fix again
4. It does not blindly overwrite files
Guarded edits use the current SHA-256 of a file.
A simple way to think about it is:
"Only change the file if it is still the same file I just inspected. If somebody changed it first, stop."
This helps prevent stale edits from overwriting newer work.
5. Shell work is supervised
PowerShell jobs have separate start, status, output, and cancel operations.
If the ChatGPT connection disappears, WorkForge does not secretly replay an old command later.
What can it do?
WorkForge currently exposes 12 MCP tools, but you do not need to memorize their names. From a user's point of view, they fit into five simple groups.
📁 Look through files and folders
"Show me the structure of this project."
"Find files related to inventory."
"Read this configuration file."
✏️ Create and edit files
"Add the new installation steps to the README."
"Refactor this name safely."
"Create a new configuration file."
🧭 Understand project state
"Read this project and tell me what I was doing."
"What has changed in Git?"
"What changed since the last commit?"
🖥️ Run PowerShell work
"Run the build."
"Run the test suite."
"Check the status of this process."
🖼️ Inspect local images
"Open this PNG and describe the UI."
"Check this image's size and contents."
What can I use it for?
WorkForge is not tied to one IDE or one game engine. If a Windows project is made of files and command-line workflows, WorkForge can often help with it.
Software and game projects
Examples include:
Unity
Godot
Node.js
Python
Web projects
CLI tools
Open-source repositories
ChatGPT can inspect the project, understand its current state, make changes, and run validation steps.
Does my project have to be a Git repository?
No. WorkForge's default mode is an ordinary local folder.
C:\Projects\MyGame
C:\Work\Prototype
C:\Documents\Notes
A folder does not need a .git directory for WorkForge to read, search, edit, inspect images, or run PowerShell commands inside it.
Git is an optional enhancement.
Without Git
→ Local Folder Mode
→ read / search / edit / images / PowerShell
With Git
→ Git Enhanced Mode
→ everything above + branches / recent commits / changed files /
staged·unstaged state / ahead·behind information
So WorkForge can install, load its profile, and work with the current folder without Git, while Git lets it understand more of the project's history.
Returning to an old project
After a few weeks away, you can say:
"Read this project and its recent Git history, then tell me where I left off."
project_resume helps inspect the current branch, changed files, and recent commits.
Debugging
"Run the build. If it fails, find the related files and investigate the cause."
This reduces the amount of error logs and source code you have to shuttle back and forth manually.
Documentation
"Rewrite the README so it matches the current implementation."
"Check whether the installation scripts and docs still agree."
Because ChatGPT can inspect both code and documentation, it can help catch stale instructions.
Repetitive local workflows
"Check these files for anything that violates this rule."
"Run the tests and summarize only the failures."
WorkForge is not a general remote-desktop robot. It focuses on files, Git, images, and PowerShell under the current Windows account and WorkForge's safety rules.
Is it difficult to install?
A normal installation takes three steps.
1. Download the ZIP
2. Extract it and run Setup.cmd
3. Connect the same Tunnel in ChatGPT
You do not need to manually hunt down Node.js and ripgrep first. Git can be added optionally during Setup if you want project-history features.
1. Download and extract
Download the latest release archive:
WorkForge-v*-win-x64.zip
Extract it to a stable local folder.
2. Run Setup
Double-click:
Setup.cmd
WorkForge checks the local components first:
Required
✓ Node.js 20+ x64 WorkForge runtime
✓ ripgrep fast file and text search
Optional
○ Git for Windows Git Enhanced Mode for branches, commits, and change history
The rule is simple:
Required component already installed → keep it
Required component missing → ask before installing with WinGet
Git missing → continue without Git or install it optionally
Node.js conflict → stop instead of layering another Node.js on top
So compatible software is not reinstalled, and missing Git does not block Setup.
Required packages use exact WinGet package IDs and --no-upgrade, so a healthy existing installation is not needlessly upgraded. Git installation is a separate choice.
The interactive Git choice defaults to Continue without Git. For automation, -InstallMissingPrerequisites covers only required Node.js/ripgrep components. Add -InstallGit explicitly if Git should also be installed.
If WinGet itself is unavailable, WorkForge does not try to bootstrap it behind your back. Setup stops and tells you to install or update Microsoft App Installer first.
The release ZIP already contains the compiled MCP server and production npm dependencies. Regular release users do not need to run npm, TypeScript, Vitest, or the repository test suite.
3. Finish the ChatGPT connection
During Setup, you need an OpenAI Platform Tunnel ID and Runtime API Key that you are authorized to use.
Then in ChatGPT:
Settings
→ Security and login
→ Enable Developer mode
→ Plugins
→ +
→ Connection: Tunnel
→ Select the same Tunnel used by Setup
Start a new chat, attach WorkForge, and use it normally.
How do I use it after installation?
You do not need a special command language.
Talk to ChatGPT normally.
For example:
"Check C:\Projects\MyGame and tell me its current state."
"Build this project and investigate any errors."
"Look at the recent work and update the README."
"Inspect this file first, then change it safely."
ChatGPT chooses the WorkForge tools it needs.
Day to day, just open WorkForge Control
After setup, you do not need to run Setup.cmd again for normal use.
Double-click:
WorkForge Control.cmd
Instead of a console menu, WorkForge now opens a local browser dashboard where you can see the important state at a glance and manage it with buttons.
Secure Tunnel Online / Offline
Health Healthy / Attention
Readiness Ready / Waiting
Supervisor Running / Stopped
Recovery Normal / Recovering
From the Dashboard you can:
- Start Tunnel so ChatGPT can reach WorkForge.
- Stop Tunnel and its supervisor safely.
- Refresh the current state immediately.
- Run Doctor to check the profile, runtime, tunnel client, credential and online path.
- Review Recent Activity in short human-readable messages.
- Open Uninstall, preview the removal with
WhatIf, and confirm before anything is deleted.
The Dashboard is not a remotely exposed admin site. It binds only to 127.0.0.1 on the current PC, uses a fresh local session each time, rejects cross-origin control requests, and shuts its background Control Server down after the browser stops making requests for a while.
The old terminal control path is still available as a recovery and advanced-user fallback:
WorkForge Control.cmd --cli
Advanced users can also invoke scripts\Control.ps1 actions such as start, stop, status, and doctor directly.
Is it safe?
WorkForge gives ChatGPT meaningful access to a workstation, so safety is part of the design rather than an afterthought.
It stays inside the current Windows user's permissions
WorkForge runs as the Windows account that launched it.
It is not a privilege-escalation tool and does not bypass Windows ACLs or UAC.
It does not install startup persistence
By default, WorkForge does not create:
Windows services
Scheduled tasks
Startup items
Run registry entries
After a reboot, the Tunnel stays stopped until the user starts it again.
It does not replay commands after a disconnect
An interrupted connection does not authorize WorkForge to replay an old PowerShell command later.
File edits are guarded against stale state
SHA-256 checks help prevent a file that changed in the meantime from being overwritten using an older version.
Runtime credentials are kept away from ordinary project commands
The Runtime API Key is stored in a protected local file and removed from the environment before project or shell code is launched.
ForgeUI logs also redact user-home paths, complete Tunnel IDs, and common credential-shaped values.
ForgeUI
WorkForge does not dump an unreadable wall of PowerShell output during setup and maintenance.
Its terminal UI shows the lifecycle as clear stages:
✓ Environment
✓ Prerequisites
✓ Runtime and profile
◆ Secure tunnel
○ Health check
○ ChatGPT handoff
Successes, warnings, failures, and next steps are easier to spot.
ForgeUI is implemented in PowerShell and does not require gum.exe or a Go runtime.
For CI, redirected output, NO_COLOR, WORKFORGE_PLAIN_UI=1, or -Plain, it automatically falls back to deterministic plain text.
What if WorkForge is already installed?
Run Setup.cmd again.
WorkForge decides automatically:
No existing profile → Install
Existing profile → Repair
Repair preserves user-edited policy files, Tunnel configuration, credentials, and related local state instead of blindly replacing them.
Upgrade also avoids overwriting user instructions. If a distributed template changed, WorkForge can place a <file>.new candidate next to the user's file for comparison.
Uninstalling WorkForge
Double-click:
Uninstall.cmd
You get two choices.
KeepWorkspace recommended
Remove WorkForge's operational connection and runtime state while keeping your workspace.
Kept
- WorkForge workspace
- Git history
- user-edited policy files
- user-created files
Removed
- Tunnel configuration
- local Runtime credential
- WorkForge runtime state and logs
- profile registry connection
- verified release engine when safe
RemoveEverything
Remove the workspace too.
Interactive mode requires the exact phrase:
REMOVE WORKFORGE
WorkForge never automatically deletes a development source checkout.
See Uninstall WorkForge for details.
What WorkForge is not
WorkForge is not:
- a remote-desktop bot that clicks anything on Windows
- a tool that secretly acquires administrator privileges
- an always-on background service
- an automation engine that blindly approves every command
- a plugin tied only to Unity or one specific IDE
Its job is narrower and more deliberate: provide a clear, verifiable working path between ChatGPT and a local Windows workspace.
Technical details
Everything below is for people who want to understand or develop WorkForge itself.
The 12 MCP tools
workstation_context
project_resume
list_directory
search_files
read_text_file
read_image
write_text_file
replace_text
shell_start
shell_status
shell_output
shell_cancel
Default profile
The default operating profile is created at:
%USERPROFILE%\WorkForge
It contains durable instructions and local profile information used while WorkForge operates.
Runtime behavior
- no Windows startup persistence is created
- Tunnel start is explicit
- unexpected Tunnel exits use bounded same-profile recovery
- disconnected commands are never replayed automatically
- PowerShell descendants are managed with Windows Job Objects
- same-profile shell work is serialized to avoid collisions
Diagnostics
powershell.exe -NoProfile -ExecutionPolicy Bypass -File scripts\Doctor.ps1 -Online
powershell.exe -NoProfile -ExecutionPolicy Bypass -File scripts\Control.ps1 -Action status
See Troubleshooting when something goes wrong.
Source development
npm.cmd ci
npm.cmd run check
npm.cmd run smoke:stdio -- workstation
npm.cmd run release
npm run check validates the TypeScript server plus no-Git installation, multi-profile loading, prerequisite detection, installation modes, the loopback-only Control Dashboard, CLI fallback, ForgeUI, Uninstall, historical Git privacy scanning, security, Tunnel recovery, and production dependencies.
Privacy gate
The public repository is scanned to prevent accidental publication of sensitive local data. The gate checks current tracked and untracked files plus reachable historical text blobs, so deleting a sensitive value from the latest commit does not make an older copy invisible.
Examples include:
personal user-home paths
non-example email addresses
real Tunnel IDs
credential-shaped values
phone numbers
private network information
runtime logs
registry and credential files
The same privacy check runs in GitHub Actions on pushes and pull requests.
Learn more
License
WorkForge is released under the MIT License.
Installing WorkForge
This server has no published package — it is built from source. Open the repository and follow its README.
▸ github.com/NotNull92/workforge-mcpFAQ
Is WorkForge MCP free?
Yes, WorkForge MCP is free — one-click install via Unyly at no cost.
Does WorkForge need an API key?
No, WorkForge runs without API keys or environment variables.
Is WorkForge hosted or self-hosted?
Self-hosted: the server runs locally on your machine via the install command above.
How do I install WorkForge in Claude Desktop, Claude Code or Cursor?
Open WorkForge on unyly.org, pick your client tab (Claude Desktop, Claude Code, Cursor) and press Install — the config is generated automatically, no JSON editing.
Related MCPs
GitHub
PRs, issues, code search, CI status
by GitHubFilesystem
Secure file operations with configurable access controls.
Memory
Knowledge graph-based persistent memory system.
Template MCP Server
A CLI tool to create a new Model Context Protocol server project with TypeScript support, dual transport options, and an extensible structure
by mcpdotdirectCompare WorkForge with
Not sure what to pick?
Find your stack in 60 seconds
Author?
Embed badge for your README
Browse similar
All development MCPs
