Connect an MCP connector to Claude without writing a line of code

If you’ve ever used Claude to look up a scientific paper or a thesis, you may have noticed that, by default, it only knows what it learned during training, plus whatever you give it in the conversation. An MCP connector changes that: it gives Claude direct, real-time access to a database or an external service, without you having to manually copy and paste search results into the chat.

MCP, in plain language

MCP stands for Model Context Protocol. Think of it as a standardized electrical outlet: instead of every service inventing its own odd plug, MCP defines a common way to connect a model like Claude to an external tool. Once the connector is set up, Claude can query that service directly during the conversation, almost as if it opened a browser tab for you and read the result.

In practice, an MCP connector can give access to open-access theses and scientific papers, to a calendar, to a codebase, or to any other service built to speak this protocol. There’s nothing magic happening on the model’s side: it’s just a standardized pipe.

On claude.ai or Claude Desktop: via a URL, no coding required

If you use Claude in your browser or the desktop app, the process requires no technical skill at all. According to Anthropic’s official help center, you go to Customize, then Connectors, click the “+” button, and choose “Add custom connector.”

You only need to fill in two things: a name so you can recognize the connector in your list, and the URL of the remote MCP server you want to connect (an HTTPS address provided by whoever hosts the service). Advanced settings exist if the service requires special authentication (an OAuth login, custom headers for an API key), but for an open connector, the URL is enough.

Once added, the connector doesn’t automatically turn on in every conversation: you have to switch it on conversation by conversation, via the “+” button at the bottom of the chat window, then checking the connector in the list.

In Claude Code: a single command line

If you use Claude Code, Anthropic’s command-line tool, the logic is the same but it happens in the terminal. The base command, as officially documented, is:

claude mcp add --transport http <name> <url>

For example, for a service accessible without special authentication, this looks something like claude mcp add --transport http my-connector https://example.com/mcp. If the service requires an authentication token, you add a header at the end of the command, like --header "Authorization: Bearer YOUR_TOKEN".

Once the command runs, the server is registered immediately, with no need to restart Claude Code. You can check that everything connected properly with claude mcp list, and remove a connector with claude mcp remove <name> if needed.

The precaution that really matters: only connect trusted sources

On this point, the official documentation is unambiguous, for both claude.ai and Claude Code: custom connectors let you link to services that aren’t vetted by Anthropic, so it’s on you to check before connecting.

In practice, that boils down to three simple things to remember. First, an MCP server that fetches outside content (a web page, a document) could, in theory, feed you back trapped content that tries to manipulate Claude’s instructions — this is called a prompt injection. Second, if the connector asks for OAuth authentication, take the time to read the permissions requested before accepting, exactly as you would before authorizing a third-party app on your Google or GitHub account. Finally, it’s better to favor connectors you already know, that someone you trust has recommended, or that come from a verified directory, rather than a random URL found in a message.

None of this is specific to MCP: it’s the same caution you’d apply before installing a browser extension or an app on your phone. The difference is that the connector acts with Claude’s capabilities, so it’s worth choosing serious sources from the start.

Key takeaways

• MCP is a standard protocol that lets Claude connect directly to an external service, in real time.
• On claude.ai / Claude Desktop: Customize > Connectors > + > Add custom connector, then fill in a name and a URL.
• The connector is then switched on conversation by conversation, via the “+” button in the chat.
• On Claude Code: claude mcp add --transport http <name> <url>, with an Authorization header if a key is required.
• Only connect connectors whose source you know: a malicious server could, in theory, try to inject hidden instructions into what it sends back.

To see this principle at work in a concrete student use case, the theses-hal connector gives direct access, from within a conversation, to French theses and the HAL open archive (France’s open repository for scholarly publications).