AI & MCP · the differentiator

Your AI can finally see what's running.

Agents guess at your runtime. Rig's local MCP hub hands them the truth, so they stop pasting kill -9 into your terminal.

Download MCP docs

95 scoped tools · Cursor, Claude, Codex, ChatGPT · hub on your computer

cursor / chat · rig mcplive

you › why is checkout 500ing?

cursor › get_logs storefront — ECONNREFUSED :5432

cursor › get_health postgres — container stopped

cursor › start_project postgres ✓

cursor › list_devices — studio-mac awake

cursor › DB was down; restarted. Retry checkout.

The gut-punch

Without Rig, your agent guesses.

It cannot see your processes, so it suggests restarts and kill commands and hopes. With Rig it reads the real state first.

Before: the agent guesses

you › why is checkout 500ing?
agent › try restarting the server
agent › maybe the port is wrong?
agent › paste this: kill -9 $(lsof -t -i:3000)
you › …and now postgres is still down
agent › I don't know what's actually running

After: Rig MCP hands them the truth

Live process state: what's up, what's crashed, who owns the port
Logs, health, and crash diagnosis against projects you already trust
Create, WordPress, Adminer, Live Link, Compose, git, env, databases, fleet
Scoped token: full, control, or read; rotate once and every editor unlinks
Source stays on your computer; the agent sees runtime, not a pasted shell

What the hub unlocks

One local hub. Every MCP client. Real runtime context.

Link Cursor, Claude, Codex, ChatGPT, or any MCP client to the hub on 127.0.0.1. They call tools against what's actually running, instead of inventing shell commands and hoping the port is free.

Check what is up

Your assistant asks. Rig answers from the live machine, not a guess.

Is the API actually running?

Yes. api (hono) is bound to :3000 and responding. The queue worker is not, it exited with code 1 six minutes ago.

Read the logs for you

Every process streamed together, with the line that matters pointed at. Wrong port, by the look of it.

[worker] booting queue listener

[worker] connecting to postgres://localhost:5433

[worker] ECONNREFUSED 127.0.0.1:5433

[worker] process exited with code 1

Know who owns the port

Which service grabbed which port, and since when. No more lsof at midnight.

:3000api

:5173web

:5432postgres

:5433nothing

Act, with your say so

Every change your assistant proposes waits for a yes. No exceptions, however nicely it asks.

Restart queue worker?

Claude wants to point it at :5432 and restart it. Nothing happens until you confirm.

Can and can't

What your assistant can and cannot do

You decide how far it reaches. Pick the mode in Settings → MCP. New installs start in Control.

Mode	What the assistant can do
Read	See projects, status, logs, ports, git history and diffs, and resource use. It can't change anything.
Control	Everything in Read, plus starting, stopping and restarting projects, freeing a port, running Docker Compose, taking a database snapshot and opening the project in a browser. No secrets, git changes, Live Links or other devices.
Full	Everything Rig can do.

It can't

Act without your yes

When it asks to do something with consequences, Rig brings its window forward and asks you. If nobody answers within three minutes, the request is refused. The agent can't approve its own request.

Start, stop, restart
Git commits, pulls and pushes
Kill a port, restore a snapshot
Live Links and other machines

It can't

Get a terminal

It only gets named actions. Git push also needs Settings → GitHub → Let Rig push, and Rig never force-pushes. A project-only token limits it to that one project.

Good to know

Your code stays put

Rig never uploads your source. But what the assistant reads through Rig (logs, a git diff) goes to the AI company that runs it. Rig hides the secrets it can recognise first.

How agents and approval work

What the agent can actually do

95 scoped tools. You choose the blast radius.

Three access modes (full, control, read) per link or per project. Rotate the token and every editor unlinks at once. Source stays on your computer; the agent sees process state, not a pasted shell.

Create & groups

Set up the morning board

Scaffold from templates, add folders, create groups, assign projects: the morning board without clicking through the wizard.

create_projectcreate_group

Run · observe · recover

Keep things running

Start, stop, restart, logs, health, port owners, crash diagnosis, make_project_work. Scoped to projects you already trust in Rig.

list_projectsget_logsget_healthwho_owns_portdiagnose_crashstart_projectrestart_project

WordPress & DB

Know the site and its data

Inventory core, themes, plugins, page builders, Woo, CPTs. Open Adminer on loopback (your computer only). Confirm-gated WP-CLI when you need a write.

wp_inspectopen_adminerdb_statuswp_rungit_status

Share · Compose · fleet

Reach beyond this project

Mint a Live Link for a client, compose_up a stack, list devices, wake a sleeping machine, screen in, or restart a catalog project remotely.

share_live_linkcompose_uplist_devicessend_device_commandwake_deviceconnect_screenbulk_restart_device

A real session

Two conversations, start to finish

The assistant checks before it acts. Here it finds a stopped database, then maps a WordPress site and shares a preview.

cursor / checkout buglive

you › why is checkout 500ing?

cursor › get_logs storefront — ECONNREFUSED :5432

cursor › get_health postgres — container stopped

cursor › start_project postgres ✓

cursor › list_devices — studio-mac awake

cursor › DB was down; restarted. Retry checkout.

cursor / WordPress sessionlive

you › what's on this WP site?

cursor › wp_inspect → Elementor + Yoast + Woo

cursor › open_adminer — loopback only

cursor › share_live_link — client preview URL

cursor › Stack mapped. Live Link ready.

Rig MCP hub linking an editor to live local processes — Local hub on your computer: scoped token, live process state, not a pasted shell.

Works with

Every assistant that speaks MCP

The hub usually lives at http://127.0.0.1:47823/mcp. That is how the editor talks to Rig, not your app's homepage. Pick your assistant for its setup guide.

One click from Rig

Open Connections in Rig and press Link. Rig writes the settings into the editor for you.

Paste the hub settings

Any other MCP client works too: copy Rig's hub settings from Connections and paste them in.

From the web, through Rig Cloud

Assistants in a browser tab can list your projects and start, stop or restart them. Your computer picks the commands up.

In the editor, through ACP

Zed and JetBrains can ask Rig Cloud about your machines, Mesh and tickets.

Connect · install

One plugin for every agent.

The Rig plugin is free and open source on GitHub. It is a single package that works in Claude, Cursor, Gemini, Copilot, Codex, Grok and Windsurf. It gives your assistant the Rig tools plus a few ready-made commands, so you do not have to set anything up by hand.

Claude Code

Claude plugin

Adds Rig to Claude Code with slash commands: /rig:setup to install and link the app, /rig:status, /rig:logs, /rig:health, and /rig:start / /rig:stop / /rig:restart, which always ask before they change anything.

claude --plugin-dir ./rig-plugin

Claude install steps

Cursor

Cursor plugin

The same plugin packaged for Cursor, with rules that tell the agent to ask Rig instead of guessing from the terminal. Or add the Rig MCP server to Cursor in one click.

Add to Cursor Cursor plugin steps

GitHub · open source

The public plugin

Rig-local/rig-plugin is the packaging only, Apache 2.0, so you can read exactly what your agent gets. The Rig app and Rig Cloud stay closed source.

gemini extensions install https://github.com/Rig-local/rig-plugin

View on GitHub

Any other assistant: one address

Paste this into any assistant that supports MCP. It signs you in to your Rig account in the browser the first time it is used, then sees the projects on the computers you have linked. Changes always need your yes.

https://api.userig.app/mcp

Codex
codex mcp add rig --url https://api.userig.app/mcp
VS Code and CopilotMCP: Add Server, choose HTTP, paste the address above.
WindsurfCascade, Manage MCPs, and put the address in the serverUrl field.

The cloud address covers your project list, health, ports and start, stop and restart. For live logs, git, databases and the rest, open Rig on your computer and use Connections, then Link, to connect the full local hub.

OpenAI · ChatGPT