nodecel
v2.3.0
Published
Deploy any local project to nodecel — the universal CLI every AI agent can call.
Readme
nodecel
Deploy any local project and get a live URL — from your terminal, or by asking an AI assistant in plain language.
Usage
Install the latest version of the nodecel CLI:
npm i -g nodecelDeploy the directory you are in:
nodecel login # authenticate this machine (opens the browser once)
nodecel deploy # packages the folder, builds it, prints the live URLThe first deploy creates the application; every deploy after that updates it. No Git repository, Dockerfile, or configuration file is required.
Commands
| Command | Description |
| --- | --- |
| nodecel deploy | Package and deploy the current directory |
| nodecel list | List the applications in your workspace |
| nodecel logs | Stream an application's logs |
| nodecel env | View or set environment variables |
| nodecel rollback | Instantly roll back to the previous release |
| nodecel whoami | Show the authenticated account and workspace |
| nodecel login / logout | Manage credentials on this machine |
| nodecel mcp install | Connect nodecel to every AI client on this machine |
Add --json to any command for machine-readable output, and --help for the
full reference. The CLI requires Node.js 18 or later and has no runtime
dependencies.
Use from an AI assistant
nodecel ships a hosted MCP server, so MCP-compatible assistants can manage your applications conversationally — no commands to remember.
nodecel mcp installOne command, every client. It authenticates once, detects the AI tools installed on your machine, and registers the server with each of them in its own config format — with the access token filled in automatically, never shown or pasted. Restart the apps afterwards so the tools load.
| Client | Where it is registered |
| --- | --- |
| Claude Code | claude mcp add (user scope) |
| Codex (CLI, IDE, desktop) | ~/.codex/config.toml |
| Cursor | ~/.cursor/mcp.json |
| Windsurf | ~/.codeium/windsurf/mcp_config.json |
| Antigravity (Google) | ~/.gemini/config/mcp_config.json (+ legacy copies kept in sync) |
| Gemini CLI | ~/.gemini/settings.json |
| Claude Desktop | claude_desktop_config.json (via mcp-remote) |
| VS Code | code --add-mcp (user profile) |
Target a specific client with nodecel mcp install codex (comma lists work),
preview with --dry-run, and remove everything with nodecel mcp uninstall.
Any other MCP client can connect to https://mcp.nodecel.com
with a bearer token from nodecel login.
On Codex, the installer also prepares the command sandbox for deploys: Codex
blocks outbound network for shell commands by default, which would dead-end
nodecel deploy mid-upload. Install enables
[sandbox_workspace_write] network_access = true and human-reviewed approvals
— additively, and never overriding a value you have already set yourself. (The
Codex CLI honors this immediately; the desktop app has a known gap, see
openai/codex#13373.)
What you can say
Once connected, just talk to the assistant in plain language:
| Say something like… | The assistant will… | | --- | --- | | "Show my nodecel apps" | list your apps with live URLs, status, and any health alerts | | "How is contextser-landing doing?" | pull traffic and error-rate analytics (24h / 7d / 30d) | | "Deploy github.com/acme/site as a React app" | create the app, provision hosting + HTTPS, and stream the first build | | "Deploy this folder" | hand your shell-capable assistant a one-time upload link — one command, and the build streams live | | "Make a landing page and deploy it" | the assistant writes the files and ships them inline — no repo, no shell, no CLI | | "Redeploy so-express-web" | trigger a deploy and stream the step-by-step progress | | "Why did that deploy fail?" | diagnose the failure and propose (or apply) a fix, then redeploy | | "Roll back to the version from before today" | review the deployment history and roll back instantly | | "Set STRIPE_KEY to sk_live_… on my-app" | update the environment variable (masked by default) | | "Show the error logs for my-app" | fetch the relevant log slice | | "Point shop.example.com at my-app" | attach the domain, hand you the exact DNS records, verify, and issue HTTPS | | "Make shop.example.com the main URL" | promote a verified domain to primary | | "Delete the old test app" | list your apps, ask which one, preview what will be lost, and delete only after you explicitly approve |
Tools
The server exposes: whoami, list_apps, get_app, deploy, redeploy_app,
deployment_status, watch_deployment, rollback_app, delete_app,
list_deployments, diagnose_deployment, get_analytics, get_logs,
get_env, set_env, set_build, previews, functions, database, cron,
projects, search, fetch, list_domains, add_domain, verify_domain,
set_primary_domain, remove_domain, and rename_app. Read-only tools are
marked as such; deploy/rollback/delete/env/domain changes are flagged as
state-changing so a client can ask for confirmation — and delete_app
additionally refuses to run until the assistant has shown a deletion preview
and the user has explicitly approved that exact app (on clients that support
MCP elicitation, that approval is a native dialog).
One deploy tool covers every way code exists — the assistant just passes
what it has and the tool routes itself: files[] for code written in the
conversation (works even in shell-less clients like claude.ai; 60 files /
3 MB), repository for a GitHub repo, local_folder for a folder on your
machine (returns a single-use 15-minute upload link and the exact
tar … | curl -T - <url> one-liner — no token leaves the gateway), or
template (landing, coming-soon, express-api) to go live with zero
files. The server also ships MCP instructions, so assistants pick the right
path without asking. The nodecel deploy CLI remains for big local projects.
More surface, where your client supports it: watch_deployment streams each
build step live via MCP progress notifications; projects retargets a session
to another workspace; search + fetch follow the deep-research contract so
nodecel works inside ChatGPT's research modes; prompt arguments and
nodecel://apps/{slug} resources autocomplete app slugs as you type; and on
MCP Apps hosts (claude.ai, Claude Desktop, VS Code) list_apps renders an
interactive in-chat dashboard. https://mcp.nodecel.com/health reports
liveness for your uptime monitor.
License
MIT
