Inspect
Read local context before acting.
Local-first / supervised / Windows power users
chatgpt-local-agent-mcp turns ChatGPT into a supervised local operator for your own PC: files, folders, browser, desktop and terminal, with visible boundaries instead of blind trust.
Clean this project folder, but do not lose anything.
Read local context before acting.
Backup and boundary-check first.
Files, browser, desktop, commands.
Report what changed and what remains.
✓workspace boundary checked
✓backup route prepared
!waiting for approval before destructive action
The missing bridge
You describe folders, paste terminal output, upload fragments and manually stitch every step together.
ChatGPT can read the local situation, propose the move, operate through the control layer and return the evidence.
Real local work
Sort, rename and prepare file operations without turning every path into manual copy-paste.
Let ChatGPT inspect a local project before it gives advice, so the answer is based on what exists.
Use browser state, screenshots and page inspection as part of the reasoning loop.
Execute local checks, capture output and bring the result back into the conversation.
Move from “here is what to click” toward supervised action across the local desktop surface.
Prefer bounded steps, recoverable changes and clear reports over invisible automation.
Do not skip this part
This is not a sandbox and not a hosted assistant. It gives ChatGPT a local control surface. That is exactly why it is useful — and exactly why supervision matters.
Start narrow. Use workspace boundaries and conservative modes first.
Backup before edits. Make recoverability part of the workflow.
Approve destructive work. Local power is not autonomous permission.
Trust evidence. Inspect logs, output and changed files.
For the impatient
Built for people who already use ChatGPT as a thinking partner and want the missing local bridge.