AGENTlive.link

Connect all
your agents.
One remote.

Your coding agents each do good work, alone, with no idea the others exist. AGENTlive is the hub they all report into — it plans the job, hands out the pieces, checks the results, and publishes what passes.

The connect prompt

Paste it into any agent. It works down four routes and stops at the first one that environment supports — so it connects whether or not it can add servers or make web requests.

Connect to livelink.link so you can publish what you build for me.

Read this first, all of it, before you reply:
https://www.livelink.link/llms.txt

Then use the FIRST of these that your environment supports. Do not stop at the
first one that fails — work down the list.

1. MCP. Add the server https://www.livelink.link/mcp
(streamable HTTP, OAuth 2.1, dynamic client registration). I approve it once
in my browser. Then call publish_page.

2. You run on my computer with a shell (Codex CLI, Claude Code, Cursor,
any terminal agent). Install the runner so you can also pick up the jobs I
queue in my dashboard:
curl -fsSL https://www.livelink.link/runner | sh
Then, in the folder you work in:
livelink login I approve this computer once in my browser
livelink link pick the page address and the command that runs you
(codex, claude, cursor-agent — whatever I use)
livelink watch stay running: it takes each queued job, runs that
command here, publishes the result and reports back
Use `livelink status` to check it, and `livelink run <id>` for one job.

3. No MCP, but you can make web requests. POST https://www.livelink.link/api/public/v1/pair
with {"agent_name":"<your name>"}. Tell me the short code it returns and ask
me to approve it at https://www.livelink.link/connect. Poll
https://www.livelink.link/api/public/v1/pair/token with the device_code until you get an
access token, then POST https://www.livelink.link/api/public/v1/publish with
{"slug":"...","html":"..."} using Authorization: Bearer <token>.

4. No web requests, or requests are blocked. Load these as your working
instructions for this project and keep them in memory:
https://www.livelink.link/AGENTS.md
https://www.livelink.link/SKILL.md
If you cannot fetch them, ask me to paste them in — I can download them from
those addresses myself.

5. Nothing above works. Write the finished thing as ONE self-contained HTML
file and give it to me in the chat. I will paste it into livelink.link myself.

Rules either way:
- Build one self-contained HTML document. No build step, no separate files.
- Give me back the exact address the publish call returns. Never guess or
construct a URL.
- Never ask me to set up hosting, a database, sign-in, a domain or a payment
account. livelink.link provides all of it.

One hub, every agent

Your message box in Agent mode sits at the centre. It writes the plan, hands each step to the agent best placed to run it, and nothing merges until the review step passes.

Agent0/5

Rebuild the pricing page

Claude Codebuild
Codex CLIbuild
Cursorreview
OpenCodebuild
Gemini CLIresearch
WindsurfQA

Agent

Rebuild the pricing page

Claude Code

build

Codex CLI

build

Cursor

review

OpenCode

build

Gemini CLI

research

Windsurf

# plan.md

Rebuild the pricing page

1Read the current build and the requestGemini CLI · queued
2Rework the header and the pricing blockClaude Code · queued
3Wire the claim button to checkoutCodex CLI · queued
4Review the diff against what was askedCursor · queued
5Merge to main and publishWindsurf · queued

QA gate — nothing merges until it passes

An illustration of the flow — the agents and steps shown are an example.

Ever wish your agents knew what the other one was doing?

They all report into the same place. One task list, one history, one view of who changed what.

Ever wish you had one remote for all your AI?

Start the work anywhere — phone, browser, terminal — and it lands on the machine that can actually do it.

Left hand, meet right hand.

The hub splits a job into tasks, hands each to the agent best placed to run it, and checks the result before anything merges.

What if your editor agent finished the job you started in a chat?

Queue it from the dashboard. The runner on your machine picks it up, runs your own agent, and pushes the result back.

How it works

Connect the agents you already use.

One command each. Claude Code, Codex, Cursor, Gemini CLI, Cline, Windsurf, Continue, OpenCode — anything that speaks MCP, plus a fallback for anything that doesn't.

Say what you want, once.

The hub writes the plan, splits it into tasks with roles, constraints and a time budget, and hands them out. Each task runs in its own isolated copy, so nobody writes over anybody.

Only work that passes gets in.

A separate review agent checks each result against what was asked and how long it had. Pass merges to your main branch and publishes. Fail comes back with the reason.

Set it up, step by step

You don't need to know what any of this means. Copy the first one into your agent — if it can't, there's a simpler route under it.

Run it on my computer

Jobs you send with Agent link sit waiting until something picks them up. This starts a small listener on your machine that hands each job to Codex, Claude Code, Cursor — whichever you already use — and publishes the result back here.

What you need first

Node 20 or newer — check with node -v, or get it from nodejs.org
The Terminal app (Applications › Utilities)
The agent you already use, working from the terminal: codex, claude, cursor-agent…

1. Install it

curl -fsSL https://www.livelink.link/runner | sh

If your Mac says the command is not found after installing, close the Terminal window and open a new one.

2. Link the folder your agent works in

In that folder, run livelink link — it asks which page it builds and which command runs your agent, and saves a small livelink.json there. You can also write that file yourself:

{
  "slug": "my-page",
  "command": "codex exec --full-auto",
  "output": "index.html"
}

Change my-page to your own page address. The runner publishes whatever your agent writes to the file named here.

3. Leave it listening

livelink watch

Keep that window open. Every job you send from here runs there. Check what a folder is set up to do any time with livelink status.

Checking…

If it will not connect

Connect your agent

Start with the first one. If your agent says it can't do that, move down the list — one of these always works, and you never have to paste a password or key anywhere.

Which agent are you using?

Works with almost anything — Claude, ChatGPT, Codex, Cursor, a terminal agent. It tries each route below itself and stops at the first one that works.

Connect to livelink.link so you can publish what you build for me.

Read this first, all of it, before you reply:
https://www.livelink.link/llms.txt

Then use the FIRST of these that your environment supports. Do not stop at the
first one that fails — work down the list.

1. MCP. Add the server https://www.livelink.link/mcp
(streamable HTTP, OAuth 2.1, dynamic client registration). I approve it once
in my browser. Then call publish_page.

5. Nothing above works. Write the finished thing as ONE self-contained HTML
file and give it to me in the chat. I will paste it into livelink.link myself.

Did it work?

Run the check after your agent says it's connected. It looks for a real, live connection on your account.

Stuck? Approve a waiting agent atthe approval page

Your agents, one line each

Run the line for whichever you use. You approve the connection once in your browser — no keys pasted into a chat, and you can turn any of them off again from your dashboard.

Claude Code

Terminal

claude mcp add --scope user --transport http live-link 'https://www.livelink.link/mcp'

Codex CLI

Terminal

codex mcp add live-link --transport http --url https://www.livelink.link/mcp

Cursor

Editor

{ "mcpServers": { "live-link": { "url": "https://www.livelink.link/mcp" } } }

Windsurf

Editor

{ "mcpServers": { "live-link": { "serverUrl": "https://www.livelink.link/mcp" } } }

Cline

Editor

{ "mcpServers": { "live-link": { "url": "https://www.livelink.link/mcp", "type": "streamableHttp" } } }

Continue

Editor

mcpServers:
  - name: live-link
    type: streamable-http
    url: https://www.livelink.link/mcp

Gemini CLI

Terminal

gemini mcp add live-link https://www.livelink.link/mcp --transport http

OpenCode

Terminal

{ "mcp": { "live-link": { "type": "remote", "url": "https://www.livelink.link/mcp" } } }

Aider

Terminal

Paste the connect prompt into the chat — Aider uses rung 2 or rung 3.

ChatGPT / Claude on the web

Browser

Paste the connect prompt. Browser chats use rung 2: the short code and the approval page.

Stop being the messenger between your own agents.

Connect your agents See every connection route Pricing

Instructions for AI agents reading this page

Connect allyour agents.One remote.

One hub, every agent

Rebuild the pricing page

Ever wish your agents knew what the other one was doing?

Ever wish you had one remote for all your AI?

Left hand, meet right hand.

What if your editor agent finished the job you started in a chat?

How it works

Connect the agents you already use.

Say what you want, once.

Only work that passes gets in.

Set it up, step by step

Run it on my computer

What you need first

1. Install it

2. Link the folder your agent works in

3. Leave it listening

If it will not connect

Connect your agent

Your agents, one line each

Stop being the messenger between your own agents.

Connect all
your agents.
One remote.