Make.com MCP Guide: Build Scenarios as MCP Tools and Call MCP Servers From Make

Make works with MCP in two directions. Your Make scenarios can become MCP tools that Claude, ChatGPT or Cursor call, through Make MCP server or a scoped MCP toolbox. And your scenarios and Make AI Agents can call outside MCP servers through the MCP Client app.

Our Make MCP server profile covers what the hosted server does and how to connect a chat app. This guide is the builder's view: how to design a scenario so an AI uses it well, which exposure option to pick, how to limit access, and how to use MCP from inside Make. Steps were checked against Make's Developer Hub and Help Center on 2 November 2026.

How Make exposes MCP

Make's Introduction to MCP names two options:

The Make MCP server docs say scenario run tools are available on all plans and management tools on paid plans. The server is Make-hosted and "runs via stateless Streamable HTTP." Connect with OAuth at https://mcp.make.com/v2, or with an API key at https://<MAKE_ZONE>/mcp/v2 and an Authorization: Bearer header.

Step 1: Build a scenario that works as a tool

Make's first-tool example walks through a "Greeting Generator." The same four parts apply to any tool:

1. Schedule it on demand. Create the scenario and set Scheduling to On demand, "so it only runs when called." Only active, on-demand scenarios show up as tools.
2. Define scenario inputs. Click the Scenario Inputs icon and add each input with a name, type and description. The AI fills these in when it calls the tool.
3. Return the result. End the flow with Scenarios > Return output and map your result into its output fields. Without it, the AI gets no answer back.
4. Write the description. In Options > Edit description, describe what the scenario does and what it needs. Make says this "will be visible to the AI as the tool's description."

Test it by running the scenario manually with sample inputs, then check that the outputs come back.

Write descriptions for the model

The model chooses tools from names and descriptions alone. Compare:

• Weak: "Customer scenario."
• Strong: "Looks up a customer by email in our CRM and returns their name, plan and last order date. Read only. Input: the customer's email address."

Say whether the scenario changes data. Our tool annotations guide explains why read-only and write tools should be clearly separated.

Step 2: Expose it

Option A: Make MCP server

Connect your AI client to Make MCP server using the profile's steps. When you connect with OAuth, you choose scopes, and "scopes determine the tools available."

If you connect with an API key, the mcp:use scope gives access to "all active and on-demand scenarios across all of your Make organizations" by default. Narrow it with query parameters:

• ?organizationId=<id> for one organization
• ?teamId=<id> for one team
• ?scenarioId=<id>, or ?scenarioId[]=<id1>&scenarioId[]=<id2> for several

The levels are mutually exclusive. These filters apply to scenario tools only, not management tools.

Option B: An MCP toolbox

MCP toolboxes are "dedicated Make MCP servers" with only the scenarios you select:

1. Click MCP Toolboxes in the left sidebar, then Create.
2. Name the toolbox and select tools. Only active, on-demand scenarios appear.
3. On the Connect tab, click Add key.
4. Copy the MCP server URL. Send the key as Authorization: Bearer <API KEY>, or append /t/<API KEY> to the URL if your client cannot send headers.
5. Optionally turn on Restrict access by IP address.

For each tool you can set a label, description and a Behavior of Read only or Read & write. Make says the default is Read & write, and that marking a tool read-only helps clients "understand how cautious to be." Set it honestly.

Toolboxes are the better choice when you hand tools to a client or contractor. You can create a separate key for each person instead of sharing one.

Step 3: Test the tool

Make's docs list the MCP Inspector as one way to see tool names and schemas:

npx @modelcontextprotocol/inspector --cli https://<MAKE_ZONE>/mcp/v2 --transport http --header "Authorization: Bearer <API_KEY>"

Add --method tools/list to print the list and exit. Check that your scenario appears with the inputs you defined. Then ask your AI client to use it with a realistic request. Our MCP Inspector lesson shows how to call tools from the CLI.

Timeouts and long scenarios

Make documents these limits for a tool call to return output:

If a scenario run times out, "the scenario continues running in Make for up to 40 minutes." The timeout response includes an executionId the client can use to fetch the result later, as long as the scenarios:read scope is granted.

Design for this. Keep tool scenarios short, or split them: one quick scenario that starts the work and returns an ID, and another that checks the result.

Step 4: Call MCP servers from Make

The other direction uses Make's MCP Client app. Make labels it "open beta," and says "both product functionality and pricing may change."

1. Add the MCP Client module to a scenario and click Create a connection.
2. In MCP Server, choose a verified server (the URL is prefilled) or + New MCP Server and enter a URL.
3. Enter an API key or access token if the server needs one.
4. If the server asks for an OAuth redirect URI, add https://www.make.com/oauth/cb/mcp.
5. Save, sign in if prompted, then pick a tool and map scenario data into its inputs.

You can also give Make AI Agents MCP tools: Make's docs describe adding an MCP connection in an agent's MCP section and toggling on the tools the agent may use. Toggle on only what that agent needs.

Pitfalls

• Scenario not showing up. It must be active and scheduled on demand.
• Empty answers. You forgot the Return output module.
• API key in the URL. It works, but URLs end up in logs and shared configs. Prefer the header form.
• Broad API keys. Without query filters, an mcp:use key reaches every on-demand scenario in every organization.
• Write tools marked read-only. Clients may skip confirmation. Use the tool permission planner to sort tools first.

When to choose something else

Make suits teams that already automate in Make. If you are comparing automation platforms for MCP, see Composio vs Zapier MCP and our Zapier MCP setup guide.

Next step

New to MCP? The free MCP course covers tools, transports and permissions. If you want Make scenarios designed as reliable MCP tools, or a custom server where Make falls short, see our MCP development service.

Frequently asked questions