Agent Stack — for Claude Code, Codex & Grok users
You already have the agent: Claude Code, Codex, or Grok — in the terminal, on the desktop, or on your phone. These are the four tools we use alongside it: one to run its work on your own machines, one to ask other models, one to find more MCP servers, and one whiteboard that the agent and you look at together.
The stack at a glance
| Tool | What it adds to your agent | How you attach it | Price |
|---|---|---|---|
| AI Commander | Shell commands and detached jobs on machines you own — server, dev box, VPS, home PC | Remote MCP: https://aicommander.dev/mcp + a host agent on each machine | Free up to 10 machines; Pro $49/month |
| PLAI.chat | A web chat with 300+ models, for a second opinion from another model | Open it in a browser; no install | Free models without login; paid models pay-as-you-go (10% top-up fee: $10 top-up → $9 credit) |
| MCPfinder | Lets the agent search for, evaluate, and get install config for other MCP servers | Local stdio MCP: npx -y @mcpfinder/server | Free, open source (AGPL-3.0-or-later) |
| cnvs.app | A live whiteboard the agent reads and draws on while you watch the same board | Remote MCP: https://cnvs.app/mcp | Free, no signup |
1. AI Commander — run your agent’s work on your own machines
AI Commander is a remote harness, not another agent. Install the small host agent on each machine you want to drive; it makes one outbound connection to aicommander.dev and accepts no inbound connections, so you open no ports. Then attach the MCP server in the client you already use, and that client can run commands on the machine, start detached jobs that keep running after the chat ends, and check their status and logs later. There is no AI Commander app: Claude, Codex, or Grok is the UI.
claude mcp add --transport http aicommander https://aicommander.dev/mcp codex mcp add aicommander --url https://aicommander.dev/mcp grok mcp add --transport http aicommander https://aicommander.dev/mcp
On the phone, open the Claude, ChatGPT, or Grok app and use the AI Commander connector where that client supports one. Setup for every client is on Connect your AI client; per-client pages: Claude Code · Codex · Grok.
Pricing: commands, detached jobs, desktop screenshots, and GPU discovery are free for up to 10 machines.
Pro is $49/month, billed monthly through Stripe, and adds file transfer (remote_pull / remote_push, files up to 100 MiB each)
and machines up to the technical ceiling of 100. No annual plan and no per-seat tier. Full details: Pricing.
More on agent-to-agent runs: Remote coding agents.
2. PLAI.chat — 300+ models in one web chat, pay-as-you-go
When you want a second opinion from a model your agent does not use — GPT, Gemini, DeepSeek, Llama, Mistral, and others — PLAI.chat puts 300+ models (via OpenRouter) in one web chat, and lets you switch models mid-conversation. Free models work with no account and no card, within a daily message limit. Paid models are billed per message from credit you add; there is no subscription and nothing renews.
3. MCPfinder — let the agent find its own MCP servers
MCPfinder is “an MCP that finds MCPs”: an open-source
(GitHub, AGPL-3.0-or-later) local MCP server that searches the
Official MCP Registry, Glama, and Smithery. It gives your agent four tools — search_mcp_servers, get_server_details,
get_install_config, and browse_categories — so it can look up a server for the task in front of it and return the config to install it.
It runs over stdio with npx (Node.js 22.15 or newer); the project’s README shows this config for Claude Code, Claude Desktop, and Cursor:
"mcpfinder": { "command": "npx", "args": ["-y", "@mcpfinder/server"] }Free. Review what it suggests before you install it — every MCP server you add acts with the permissions you give it.
4. cnvs.app — one board for agent and human
cnvs.app is a free, no-signup, real-time whiteboard in the browser with an MCP endpoint at
https://cnvs.app/mcp. Through it, your agent can read a board, add text, links, images, and freehand strokes, move and erase items, wait for your
edits, and manage kanban columns and tasks — on the same live board you have open. Its edits appear for everyone on the board, tagged with an
ai: author label so you can tell them apart. Documented clients include Claude Code and Claude Desktop
(cnvs.app MCP docs):
claude mcp add --transport http cnvs https://cnvs.app/mcp
There is no API key or sign-in: the board ID is the key, so treat a board link like a password and do not put secrets on a shared board.
How the pieces fit
- Plan on the board. Sketch the task on a cnvs.app board; your agent reads it and adds a kanban card per step.
- Find the tools. If a step needs an integration, the agent asks MCPfinder for a server and shows you the install config.
- Run it where the work lives. The agent starts a detached job on your own machine through AI Commander — the repo, toolchain, and GPU are already there.
- Check from anywhere. From your phone, ask the same client for the job’s status and logs; get a second opinion on the result from another model in PLAI.chat.
FAQ
Start with your own machine
Install the host agent on a machine you own, attach https://aicommander.dev/mcp in Claude Code, Codex, or Grok, and ask for a real job on that box.
Disclosure: all four products are made by the Coder AI team. Prices are as published on each product’s own site on 2026-10-08; check there before you buy. Claude and Claude Code are trademarks of Anthropic; Codex and ChatGPT are trademarks of OpenAI; Grok is a trademark of xAI. AI Commander is not affiliated with, endorsed by, or sponsored by these companies.