Connect an AI Assistant (MCP)

Liinks runs an MCP server. MCP, the Model Context Protocol, is the standard AI assistants use to connect to outside services, and connecting one to Liinks means you can describe a change in plain language instead of opening the editor. "Add my new podcast episode to my page." "Move the newsletter signup to the top." "What did people submit through my contact form this week?"

This is built on the same Public API, so anything the assistant does is something you could have done with a REST request yourself.

What you need

1. A Liinks account. Any account will do. There is no plan requirement for connecting an assistant.
2. An MCP-capable client. Claude, ChatGPT, Cursor, Meta Muse, and a growing number of other assistants and editors support MCP.

Connect your client

The Liinks MCP server lives at https://www.liinks.co/api/mcp and uses the Streamable HTTP transport. In ChatGPT you don't need the address at all, because Liinks is listed in its app directory. Other assistants connect by URL, or with an API key.

ChatGPT

1. Open the app directory at chatgpt.com/apps and search for Liinks.
2. Click Connect. ChatGPT opens the Liinks consent screen described below. Sign in if you need to, check the permissions, and click Allow access.
3. Start a chat, pick Liinks from the + menu, and ask away.

If Liinks doesn't show up in the directory for your account, add it by URL as described next. In ChatGPT, that means turning on developer mode first, under Settings > Apps & Connectors > Advanced settings.

Other assistants: sign in by URL (recommended)

Most clients let you add a remote MCP server by URL alone. Look for a settings screen named Connectors, MCP servers, or Integrations, choose the option to add a custom or remote server, and give it:

https://www.liinks.co/api/mcp

That is the whole setup. The first time the assistant tries to use Liinks, it sends you to the Liinks consent screen: you sign in as normal, review what the assistant is asking for, and approve it. Nothing is copied or pasted, and no key ever sits in your client's configuration.

The consent screen

Whether you connect from ChatGPT's directory or by URL, the consent screen names the client and shows exactly which permissions it wants:

• See your page: display name, bio, links, blocks, theme, and tabs.
• Edit your page: add, change, reorder, hide, or remove blocks, and update your bio, socials, and theme.
• Read your audience data: form submissions and email subscribers.
• Read your analytics: page views, clicks, top links, and traffic sources.

Approve only what you are comfortable with. A connection cannot grant itself more later: if an assistant needs a permission you did not give it, it has to ask you again.

Connections appear under Settings > API Keys as Connected app rows alongside any keys you made yourself. Revoking one there ends that assistant's access immediately.

Paste an API key instead

Some clients, and most local editor setups, still want a static token in a config file. That works too. Create a key from Settings > API Keys (see Public API for the walkthrough), then paste the JSON your client expects:

{
  "mcpServers": {
    "liinks": {
      "type": "http",
      "url": "https://www.liinks.co/api/mcp",
      "headers": {
        "Authorization": "Bearer liinks_live_<your-token>"
      }
    }
  }
}

Replace liinks_live_<your-token> with your own key, keeping the word Bearer and the space in front of it.

Where that JSON goes depends on the client. Look for a settings screen named MCP servers, Connectors, or Integrations, and check your client's own documentation for the exact file or field. Whatever the wrapper looks like, Liinks only needs those two things: the endpoint address and the authorization header.

An API key is not scoped the way a signed-in connection is. It can do everything your account can do through the API, so treat it like a password.

Once it is connected, ask the assistant to list your Liinks profiles. If it comes back with them, the key and the connection are both working.

What the assistant can do

The server exposes fourteen tools. Your assistant picks the ones it needs, so you do not have to name them, but it is worth knowing the shape of what you have handed over.

Reading

• list_profiles lists the profiles your key can manage.
• get_profile reads one profile in full: its fields, styles, social icons, blocks, and tabs.

Editing the profile

• update_profile changes your display name, bio, tagline, or slug.
• set_styles changes the theme: colors, fonts, background, and layout.
• set_socials replaces the row of social icons.

Editing blocks

• add_block adds a link, text, divider, folder, or media block.
• update_block edits a block, or hides and shows it.
• remove_block deletes a block.
• reorder_blocks sets the top-to-bottom order of blocks on a tab.

Reading your data

• list_form_submissions reads form submissions, newest first.
• list_subscribers reads email subscribers, newest first.

Reading your analytics

• get_analytics reports views, visitors, clicks, and click-through rate for a period, with the previous period for comparison and your top blocks and social icons.
• get_analytics_breakdown groups that traffic by country, referrer, device, campaign, and more.

Checking the result

• screenshot_profile renders your live page to images so the assistant can look at what it just did.

Block types that only the editor can create, such as forms and subscriber blocks, are not in that list. Add those from the editor, then let the assistant reorder or edit them like any other block.

Before you let it write

A few things are worth knowing up front.

Changes go straight to your live page. There is no draft step and no preview. If you want to see something before it lands, ask for it: "show me what you would change first" works, and so does asking for a screenshot afterwards.

Deletions ask first. remove_block is marked as destructive, and the tool tells the assistant to confirm with you before it runs. Well-behaved clients will stop and check. Treat that as a courtesy rather than a guarantee, since how strictly a client honors it is up to the client.

The assistant gets exactly what the key gets. It can reach the profiles that key can manage and nothing else. It cannot see your password, your billing, or your account settings.

You can cut it off instantly. Revoke the key or the connected app from Settings > API Keys and it stops working immediately. Do that if you ever stop using a client, or if you pasted the key somewhere you are not sure about.

It shares your API rate limits. 120 requests per minute and 30 writes per minute, the same budget as the REST API. A long editing session will not come close, but a script running alongside it might.

The developer portal

liinks.co/developers is the reference page for all of this: the setup steps above, a REST quickstart, the error model, current rate limits, and the versioning policy. It is also where the machine-readable resources live, including the OpenAPI specification and the MCP manifest.

The API is in beta, so shapes may still change. We will give notice before anything breaks.

Need help?

If your client will not connect, or a tool does something you did not expect, contact our support team and tell us which client you are using.