How to Connect Claude to Make MCP Toolboxes

Connecting Claude to Make MCP Toolboxes is now easier than it was when remote MCP setups required manual URL and token entry. Make is now available as a built-in connector in Claude, which means many teams can connect through OAuth instead of assembling the connection by hand.

The core idea has not changed: a Make MCP Toolbox exposes a selected set of scenarios as MCP tools. Claude can then call those tools when the task fits. The difference is that the connection path is smoother, especially if your goal is to let Claude trigger a curated set of workflows instead of giving it broad workspace-level access.

What you will build

You will create or reuse a Make MCP Toolbox, add the scenarios you want Claude to use, configure tool metadata, and connect the toolbox to Claude. In many cases, the fastest route is to use Make’s built-in Claude connector. If that option is unavailable in your environment, you can still use the standard remote MCP server method with the toolbox URL and key.

When to use a toolbox instead of the full Make MCP server

Choose a toolbox when you want Claude to access only a limited set of scenarios. This is the safer and cleaner option for use cases like “draft an email from a CRM record,” “check a support queue,” or “create a follow-up task after summarizing a note.”

Choose Make MCP server when you need broader access to your Make account and are comfortable managing larger permission scopes. A toolbox is the better fit for most operational assistants because it reduces accidental tool sprawl.

What you need before you start

• A Make account with scenarios prepared as MCP tools.
• Those scenarios must be active and use on-demand scheduling.
• A Claude account or workspace that supports the Make connector or remote MCP server connections.
• Clear tool boundaries so Claude can choose the right scenario without ambiguity.

Step 1: Prepare your Make scenarios

Before you connect anything to Claude, simplify the scenarios you want to expose. Good MCP tools are narrow and well-labeled. A tool like “Find deal by company name” or “Create follow-up task in Asana” is easier for the model to use than a compound scenario with multiple unrelated branches.

Also decide which actions should stay read-only. If a workflow only reads a CRM record or checks a row in Google Sheets, keep it separate from workflows that write or delete data.

Step 2: Create the Make MCP Toolbox

In Make, create a toolbox and add the prepared scenarios. The toolbox will generate its own URL and key. You can issue more than one key for the same toolbox if you want to separate production access, shared client access, or testing access.

As part of the toolbox setup, customize the tool names and descriptions. Claude is more likely to choose the correct tool if the name says what the tool does and the description says when it should be used.

Step 3: Set behavior annotations carefully

Make lets you label tools as read only or read & write. Use that. These annotations help MCP clients understand how cautious they should be with a tool. They are not just documentation; they shape how comfortably the client can invoke the tool.

For example, “Look up unpaid invoices” should be read only. “Create invoice reminder draft” might be read & write, depending on how the scenario is designed. Separate them instead of trying to cover both jobs in one tool.

Step 4: Connect Make to Claude

If your Claude environment supports the new built-in Make connector, go to Claude’s connector settings, browse connectors, select Make, and complete the OAuth flow. This is the cleanest option because it avoids manual URL and token entry.

If you are using the generic remote MCP flow instead, configure the toolbox URL based on what the client supports. Make recommends HTTP streamable transport, especially the stateless path, for better reliability. If your client supports authorization headers, pass the key there as a Bearer token. If it does not, use the URL structure that includes the key.

Step 5: Test with one simple request

Start with a single safe action. Good first tests are looking up a CRM record, fetching an internal document, or creating a draft instead of sending a message. Confirm that Claude can discover the tool, call it once, and handle the returned data correctly.

Then test one write action in a sandbox. If the wrong tool gets called, the problem is usually not the connection. It is usually weak naming, poor descriptions, or overlapping tool responsibilities.

How to verify the workflow is working

• Claude can see the connected Make tool or connector.
• A request triggers the expected Make scenario.
• The scenario output comes back in a format Claude can use.
• Read-only tools stay read-only in real use.

Common problems and fixes

Claude is connected but not using the tool

Rewrite the tool name and description in the toolbox. Be specific about inputs, outputs, and the situations where the tool should be called.

The scenario is missing from the toolbox

Check that the scenario is active and set to on-demand scheduling. Ineligible scenarios do not appear as toolbox tools.

The action takes too long

Toolbox scenario runs time out after 40 seconds. Shorten the workflow, reduce branching, or move long-running work to an asynchronous follow-up scenario.

The connector works in test but behaves inconsistently later

This often means the tools overlap too much. Keep one tool for lookup, one for drafting, and one for state-changing actions instead of combining everything.

When a template helps

A Make template is useful when the scenario pattern is already common, such as CRM enrichment, support classification, task creation, or Google Sheets updates. It saves time on module wiring, but it does not save you from MCP-specific work like naming tools well, limiting permissions, and checking outputs.

FAQ

Is the built-in Claude connector better than manual MCP setup?

Yes, if it is available in your environment. It reduces setup friction and avoids manual URL and token handling.

Can Claude use multiple tools from one toolbox?

Yes. A toolbox is a collection of selected Make scenarios exposed as MCP tools.

Should I connect Claude to the full Make MCP server instead?

Only if you need broader access. For most assistant-like workflows, a toolbox is easier to govern.

What should I expose first?

Start with read-only lookup tools and low-risk drafting flows. Add write tools after the basic tool selection behavior is reliable.

Final notes

The connection itself is the easy part. The harder and more important work is deciding which workflows Claude should be allowed to use and describing those workflows precisely enough that the model can pick the right one. If you get tool boundaries right, Make and Claude work well together.