How To Add A Claude Custom Connector From An MCP URL

Someone gave you an MCP URL and said "just add it to Claude". You opened Settings, looked around, and found nothing called MCP anywhere.

That's because Claude doesn't call it MCP in the menus. It calls it a connector, and a URL someone hands you becomes a custom connector. Once you know where the button lives, adding a Claude custom connector takes about a minute and works on every plan, Free included.

The fiddly part is the screen after you paste the URL. It asks how the connector should sign in, offers three options with three more underneath, and gives you no hint which one the server wants.

I'll go through the whole thing in order: where the menu is, what each sign-in choice means, how to switch the connector on in a chat, and the three problems people hit afterwards. The worked example is a publishing connector, because it's a good test. If it works, the chat can put a page online and hand you a link, which is easy to check.

What Is A Claude Custom Connector?

A Claude custom connector is a remote MCP server you add to Claude yourself, by pasting its URL. MCP, the Model Context Protocol, is the open standard that lets an AI assistant call outside tools. The Model Context Protocol's own introduction A server publishes a list of tools, and once it's connected, Claude can call them when your request needs one.

Claude ships with a directory of ready-made connectors for big services. A custom one is anything that isn't in the directory: a tool a developer built, a small company's server, or a host's publishing service.

The reason it matters is simple. A chat on its own can read the web but can't do anything out there. A connector is how it gets hands. If you want the background on what that means for publishing, what a Claude chat can and can't host covers it.

Which Claude Plans Can Add One?

All of them, with two catches. Anthropic's help article on custom connectors using remote MCP, which I read in November 2026, lists Free, Pro, Max, Ultra, Team and Enterprise, across Claude on the web, Cowork and the desktop app.

On Team and Enterprise, members don't add the URL themselves. An owner adds it in organisation settings, and members then press Connect on their own Connectors page.

If an older guide tells you custom connectors aren't on the free plan, go by Anthropic's current article instead. It lists Free.

How Do You Add A Claude Custom Connector?

On Free, Pro, Max or Ultra, it's eight short steps:

1. Open Customize, then Connectors.
2. Press + Add, then Add custom connector.
3. Type a name and paste the server's URL, then press Continue.
4. Review the sign-in settings Claude detected, and press Continue.
5. Under Authentication, pick Sign in now, Sign in when needed, or No sign in.
6. If you picked a sign-in option, choose the OAuth client. Go with what the server's docs say, or what step 4 detected.
7. Under Request headers, add a key or token only if the server's docs tell you to.
8. Press Add.

On Team or Enterprise, the owner does the same from Organization settings, then Connectors, choosing Add, then Custom, then Web.

The URL has to be a remote server, the kind that lives at an https:// address. A server that only runs on your own computer can't be added this way. That kind belongs in a desktop app or a coding agent like Claude Code.

Which Sign-In Option Should You Pick?

Pick the one the server was built for, and if its docs don't say, the screen in step 4 usually does. Claude reads the server's settings and shows what it found.

The OAuth client choices under the sign-in options are for developers, and the right one depends on how the server was built. Claude marks one as recommended. If signing in fails with it, remove the connector, add it again and pick Register automatically, which suits servers that let apps register themselves.

My rule: if a server offers sign-in, use it rather than pasting a key into a header. A sign-in can usually be revoked from the service's side in one click, and you never have to copy a secret around.

A Walkthrough: A Publishing Connector

Here's the whole thing with a real server. Say you run a volunteer running club and want Claude to turn the weekly run list into a schedule page and put it online, without making an account anywhere first.

Add a custom connector named OxyPages with this URL, and pick No sign in:

https://mcp.oxypages.com/mcp

Then open a new chat, press + at the bottom left, go to Connectors, and switch it on. Paste the run list and write: "Make this a simple schedule page and publish it with the OxyPages connector."

Claude builds the page, calls the connector's publish tool, and replies with two links: a live page and a claim link, with a note that the page goes offline in 30 minutes unless you claim it. That's the same no-account pattern coding agents use, explained in how to deploy without an account. A club page is personal, non-commercial use, so claiming it into a free account is enough. Share the new address the claim gives you.

To update the page from the chat the next week instead of starting over, add a second connector with the account address, pick Sign in now, and choose Register automatically under OAuth client, because this server lets apps register themselves:

https://mcp.oxypages.com/mcp/account

That sends you to the host's site to log in and approve. From then on the chat can list your sites and publish new versions, but not delete a site or touch billing, domains or your account. On a free Claude plan you'd have to remove the first connector to add this one, because of the one-connector limit.

How Do You Switch It On In A Chat?

Press + at the bottom left of the chat box, choose Connectors, and flip the toggle beside it. The setting is per conversation, so a new chat may need it switched on again.

Then name it in your request, using the name you gave it. "Publish this with my publishing connector" works better than "put this online", because Claude knows exactly which tool you mean.

Why Isn't My Connector Working?

Three problems cover almost every case.

It Doesn't Show Up In The Chat

It's switched off for this conversation. Check the + menu under Connectors. On Free, check you haven't already used your one custom connector on something else. On Team or Enterprise, an owner has to add it before you can connect.

The Sign-In Keeps Looping

The service's login is failing quietly, then Claude asks again. Allow pop-ups for claude.ai, finish the login in the same browser, and check the service doesn't want something first. Many want a verified email, and an account with two-factor sign-in will ask for the code.

Claude Answers But Never Calls The Tool

Claude decided it could answer without the tool. Name the connector in your request, and if Claude asks permission to use the tool, allow it. If it still won't, remove and re-add the connector so Claude reloads the server's tool list.

One honest note on publishing connectors with no sign-in: they're rate limited, and chat users share network addresses, so a busy hour can mean a refused publish. Try again a few minutes later, or publish the Claude HTML file yourself. If you need a page that runs server code or a database, no publishing connector will help, because this one and hosts like it serve static files only.

What To Do Next

Pick a connector you actually need, and add it in the next five minutes while the steps are fresh.

1. Find the server's URL in its docs.
2. Add it under Customize, then Connectors.
3. Pick the sign-in option the server expects.
4. Switch it on in a new chat and name it in your request.

If you're working in Claude Code rather than a chat, you don't need a connector to publish at all. Putting a Claude Code website online shows the one-request version.

To try a publishing connector with no account, copy the connector address from our MCP page and paste it in. If you want the page to stay up, start a free account, no card needed, for personal pages. A business page needs a paid plan.