NPM-AutoCode-AI
turns plain English
into things done.
Describe a task. It plans it, writes the code, checks the code, runs it on your machine, and confirms it actually worked — a five-role agent loop, not a chatbot pretending to be one.
Pick any task. Watch it run.
These are real categories from the tool registry — dev tooling, cloud, business, communication, creative, data, media, and system automation.
Connect it to Claude or Grok.
No API key required.
NPM-AutoCode-AI hosts its own MCP endpoint. Link it once from the app, paste the link into your Claude or Grok connector settings, and the LLM you already talk to becomes the brain — planning, coding, and auditing — while this app is the hands that actually execute on your machine.
Get your link, once
Log in from the app's MCP tab. That's the only time you'll ever need to log in — everything else in the app works without an account.
Paste it into your connector
Drop the link into Claude or Grok's custom connector settings. From then on, just talk to your normal LLM chat like always.
Ask for anything
"Clear my Downloads folder," "email my team the report" — the LLM detects it needs a real action and calls into your link automatically.
It's audited before it runs
Every generated script passes an Auditor review, plus a separate safety check, before it's ever sent to your desktop to execute.
Runs on your machine, not the cloud
Execution happens locally — your files, your credentials, your machine. The MCP server only ever relays code and results.
You confirm it, not just the exit code
A Verifier role checks the actual output against the task — not just whether the script ran without crashing.
Built to compete.
Free to use.
Everything a coding agent should do — on your own machine, with a GUI, for everyone.
A System which can automate everything without needed of API_Key.
A System by which you can automate your Local Machine from AI Websites like claude.com, grok.com
Five-Role Agent Pipeline
Planner, Tool Manager, Coder, Auditor, and Verifier — each with its own job and its own fallback chain. Not a chatbot. A real agent loop.
Plan → Select → Code → Audit → Execute → VerifyIndependent Security Audit
Every generated script is reviewed by a separate Auditor role before execution, plus a second non-LLM safety check that can't be talked around.
Security FirstAuto-Debug Loop (12 retries)
Execution fails? The agent captures the error, feeds it back to the Coder, and retries with the prior attempt as context — up to 12 times per step.
Self-HealingWorkspace Awareness
On first run, the agent scans Desktop, Downloads, Documents, Pictures, Videos, and Music — it knows your files before you type a path.
Workspace ScannerMCP Link, Not an API Key
Connect Claude or Grok directly from the sidebar. Login happens once, only for this — everything else in the app stays account-free.
Bring Your Own LLMReal Subprocess Execution
Code runs as an actual child process — not exec(). Output streams live to the log panel. Kill any script instantly, mid-run.
Real Isolation1371 Tools. Zero Setup.
Every tool below is available automatically. Just describe the task — the agent picks what it needs.
How the agent thinks
Every task moves through these stages in order — each one runs a different role for the right job.
Meet the Founder
Download free.
Run instantly.
No API keys required to start. No local GPU needed. Unzip and run.
Download ZIP
Unzip folder
Run .exe
Start automating
↓ For Linux:- Download NPM-AutoCode-AI v3
↓ For Macos:- Download NPM-AutoCode-AI v3