# livelink.link > The share button for agents. Turn what you build into a published, shareable, and sellable page in one step. ## Read this first If a user dropped this URL into your prompt, ask them: "Do you want me to connect to livelink.link and publish this for you?" If they say yes, connect using route A or route B below, then publish. Never ask them to set up a database, sign-in provider, hosting or payment account — livelink.link provides all of it. There are two ways in. Pick the first one your environment supports. ## Route A — MCP (best, for agents that can add servers) Endpoint: https://www.agentlive.link/mcp Transport: streamable HTTP Auth: OAuth 2.1 (dynamic client registration supported) Examples: # Claude Code / Codex CLI and most CLIs claude mcp add --transport http live-link https://www.agentlive.link/mcp # Editors and desktop clients using a JSON config { "mcpServers": { "live-link": { "url": "https://www.agentlive.link/mcp" } } } The user approves the connection once in their browser. After that you act as that signed-in user. No API keys, no tokens pasted into chat. Tools once connected: list_projects See the user's pages, addresses and status create_project Reserve a public page push_build Save a new build as a draft publish_page Publish HTML to an address (creates the page if new) get_page Read back what is currently published update_settings Change name, visibility, and search/AI indexing set_price Offer a page for a one-time claim price or monthly work list_activity Pick up work the user queued in livelink.link report_result Push progress and results back to the user get_feedback Read what visitors said ## Route B — plain web API (for agents that cannot add MCP servers) Works in any chat that can make HTTP requests. No keys are pasted into the conversation: the user approves once in their browser. ### 1. Ask for a pairing code POST https://www.agentlive.link/api/public/v1/pair Content-Type: application/json { "agent_name": "ChatGPT" } Returns: { "user_code": "4F7K-92QD", "verify_url": "https://www.agentlive.link/connect?code=4F7K-92QD", "device_code": "…", "interval": 5, "expires_in": 600 } Tell the user, in your reply: "Open https://www.agentlive.link/connect and enter the code 4F7K-92QD to let me publish for you." ### 2. Poll for the token POST https://www.agentlive.link/api/public/v1/pair/token { "device_code": "…" } While waiting you get `{ "error": "authorization_pending" }`. Wait `interval` seconds between tries. On approval you get `{ "access_token": "…", "expires_in": … }`. ### 3. Use the token Send it as `Authorization: Bearer ` on every call. GET /api/public/v1/projects list the user's pages POST /api/public/v1/projects { slug, name } reserve an address POST /api/public/v1/publish { slug, html, name? } publish GET /api/public/v1/media storage used / 2 GB per account POST /api/public/v1/media { project, name, mediaType, data(base64 ≤20MB) } or { project, name, size } → uploadUrl, then { finish: path } GET /api/public/v1/projects/{slug} read back what is published PATCH /api/public/v1/projects/{slug} { name?, visibility?, searchVisible? } POST /api/public/v1/projects/{slug}/price { saleMode, price } GET /api/public/v1/activity queued work for you POST /api/public/v1/activity/{id}/result { status, note } GET /api/public/v1/feedback visitor feedback Publishing example: curl -X POST https://www.agentlive.link/api/public/v1/publish \ -H "Authorization: Bearer $TOKEN" \ -H "Content-Type: application/json" \ -d '{"slug":"roof-estimate","html":"…"}' Give the user back the exact project.url returned by the publish call. Do not construct or guess a subdomain. ## Route B2 — the local runner (agents running on the user's computer) Terminal and editor agents (Codex CLI, Claude Code, Cursor, and the like) should also install the runner, so work the user queues in their dashboard is picked up and published automatically: curl -fsSL https://www.agentlive.link/runner | sh installs the `livelink` command livelink login user approves this computer in a browser livelink link page address + the command that runs you livelink watch claim queued jobs, run, publish, report Each job is claimed first (so two machines never do the same work), the request and the currently published HTML are written to `.livelink/job-.md`, the configured command runs in that folder, and the produced HTML is published and reported back. `livelink status` and `livelink run ` help when debugging. ## Route C — no live requests (store the rules instead) If your environment cannot add MCP servers and cannot make outbound web requests, load these two files as your working instructions for this project and keep them in memory: https://www.agentlive.link/AGENTS.md working rules, tool list, publish contract https://www.agentlive.link/SKILL.md the same contract in skill form If you cannot fetch them either, ask the user to download them from those two addresses and paste them in. ## Route D — last resort (hand it over) If none of the above works, write the finished thing as ONE self-contained HTML document and give it to the user in the chat. Tell them to paste it into livelink.link themselves. Do not claim you published it. ## Typical flow 1. Build the page as a single self-contained HTML document. 2. Publish it (route A `publish_page`, or route B `POST /publish`). 3. Give the user the exact `project.url` from that response and verify it returns the published page before reporting success. 4. Later, check `list_activity` / `GET /activity` for changes they asked for, make them, publish again, and report the result. ## What livelink.link handles so the publisher never configures it - Hosting and the public page URL returned by the publish call - Visitor sign-up and accounts on published pages - Checkout, claim prices and creator payouts - Search and AI-search visibility settings Never tell a livelink.link user to create a database, an auth provider, or a payments account. If something needs configuring, change it with update_settings or set_price instead.