Make MCP server and Claude connector: what it does, access, how to connect

Make hosts one official MCP server, also a built-in Claude connector, that runs scenarios and can create, change and delete Make objects. Sign in with OAuth or a token. Verified 21 Oct 2026.

Official remote MCP · read & write
Claude connector · read & write

Hosting: Remote only: Make hosts it over Streamable HTTP at https://mcp.make.com (OAuth) or your zone's /mcp endpoint (MCP token).
Auth: OAuth with an organization and scope picker, or an MCP token from your Make profile (Bearer header or in the URL).
Read / write: Read and write: scenario runs act at once; management tools create, update and delete Make objects. Claude's listing shows 67 tools.
Works with: Claude.ai, Claude Desktop, Claude Code, ChatGPT (developer mode), Cursor and other MCP clients, per Make. Make's ChatGPT plugin is unlisted for now, per Make.
Plan requirement: Scenario-run tools on all Make plans; management tools need a paid Make plan, per Make. Limiting OAuth access by team needs a Make Teams plan or higher. Make doesn't state a Claude plan for the connector.
Last changed: 2026-09-29 (Make says its ChatGPT plugin is unlisted while it works with OpenAI)
Verification: Verified 21 Oct 2026 by theFinanceStack · confidence: high

What can an agent do with Make through MCP?

It runs your active and on-demand Make scenarios as tools and, on paid plans, manages the account; Claude's directory lists 67 tools.

Tool types

Type	What it does
Scenario-run tools	Turn each active or on-demand scenario into a tool. The agent sends inputs, Make runs the scenario and returns its outputs.
Management tools (paid plans)	View and modify scenarios, connections, webhooks, data stores, teams and organizations.

What happens when a scenario runs

A run does whatever the scenario does in its connected apps.
So one that creates an invoice, posts a journal or pays a bill does so at once.
Make's MCP layer has no draft step.

Good to know

The 24 tools visible in Claude's listing include scenario create, delete, activate and run, and data store record create, replace and delete. We could not see the other 43.
No listed tool is a payment tool. Any exposed scenario that moves money becomes one.

Can

Run active and on-demand scenarios with inputs and outputs (all plans)
Create, update, delete, activate and deactivate scenarios (paid plans)
View and modify connections and webhooks (paid plans, per Make)
Create, update and delete data stores and records (paid plans)
View and modify teams and organizations (paid plans, per Make)
Fetch a timed-out run's output, with the scenarios:read scope

Cannot

Run scenarios that are neither active nor on-demand (per Make)
Limit management tools with the token's organization, team or scenario filters (per Make)

What access does it get?

Whatever scopes you tick for one organization at sign-in; Make warns that its "Run your scenarios" scope covers every active and on-demand scenario in that organization.

Where	Scopes and permissions (as documented)	Admin requirement
Make MCP server via OAuth (remote, https://mcp.make.com)	Chosen with the organization at sign-in. Named by Make: "Run your scenarios" (all active and on-demand scenarios in the organization) and scenarios:read (late outputs). Other scope names are not documented by the vendor.	Make doesn't document an admin step. To limit reach, Make suggests a Teams plan or higher with users assigned to specific teams.
Make MCP server via MCP token (remote, your zone's /mcp)	Token scopes picked in Profile › API access; mcp:use exposes scenarios in every organization the token reaches. Adding one of organizationId, teamId or scenarioId to the token URL narrows scenario tools only; management scopes on the token are not narrowed (per Make).	Make doesn't document an admin step; users create tokens in their own profile.
Make connector in Claude (built in)	Same OAuth consent: choose the organization and its scopes, then Allow.	Claude Team/Enterprise: an Owner adds the connector before members can connect.

How do I connect it?

In Claude, add Make from the connector directory and pick your organization and scopes; in other clients, add https://mcp.make.com and complete the same OAuth consent.

In Claude, open Customize › Connectors › Discover (or claude.com/connectors/make), search for Make and click +.
On Make's consent screen, choose the Organization and tick only the scopes you need, then click Allow.
Test with a read-only request before running a scenario. Listing scenarios needs a paid Make plan and management scopes; on a free plan, ask which Make scenario tools are available.
Team/Enterprise: an Owner must add the connector before members can connect.

The Claude directory lists Make as Anthropic verified, added January 2026.

What should a finance team watch for?

Scenario runs act immediately in every app the scenario touches, the main run scope is organization-wide, and long runs carry on after the tool call times out.

Nothing waits as a draft. Scenario runs execute at once, and management tools can delete scenarios or records, or activate a scheduled scenario. Keep your client's per-call approval on.
Money can move indirectly: an exposed scenario that pays, posts or sends in a connected app will do so when run. Expose only scenarios built for agent use, ideally ending in a draft.
Timeouts: run tools stop waiting after 25 seconds (OAuth) or 40 (token), management tools 30 or 60, but the run continues for up to 40 minutes. Check the execution before retrying.
Make's MCP docs don't say how agent-triggered runs count toward your Make plan's usage. Check your usage in Make after the first runs.
Least privilege: untick OAuth scopes you don't need. For a token, give it mcp:use and no management scopes, then add organizationId, teamId or scenarioId to the token URL (one level only). Make says these filters skip management tools.
Make calls the MCP token a secret. Make documents the scenario filters only on the token-in-URL form, so a pinned token sits in the URL. Use the Authorization header when you don't need a filter, and keep config files holding either form out of shared folders and repos.
Prompt injection: scenario outputs and data store records can contain text. An injected instruction could push the agent to run a paying or posting scenario.
Make's MCP docs make no data-handling statement; read Make's privacy policy and your AI provider's terms.
Make keeps a legacy local server (@makehq/mcp-server) for advanced cases only.

How could finance teams and accounting firms use it?

Reporting and lookup scenarios suit questions; scenarios that create records suit drafting, with a person approving each run that changes a system of record.

Inside a company

Run a scenario that returns overdue invoices from your accounting app before the AR call.
List scenarios and whether each is active (paid plans), then check in Make which ones touch the ledger before a close or audit.
Trigger a scenario that creates a supplier bill as an unposted draft for a person to approve.
Look up a vendor mapping held in a Make data store (paid plans).

At an accounting firm

Give a token mcp:use only and pin its URL to one client's reporting scenario. Leave management scopes off, because Make says the pin doesn't limit them.
List client scenarios that are inactive (paid plans), then review them in Make.
Run a scenario that drafts missing-receipt requests for a manager to send.
Outline a new scenario with the agent, then build and test it in Make before activating.

Whatever the use, a person should approve each run that changes the books or another system of record, and check the result in that system.

Want Make connected to Claude for your team?

We configure scoped connections and a review step before anything changes your books. Fixed fee.

Book a 30-minute setup call

Change log

Date	Change	Source
2026-09-29	Make says the ChatGPT plugin is unlisted, with temporary beta access only for earlier installs; developer-mode MCP remains	community.make.com/t/feature-spotlight-make-plugin-for-chatg
2026-09-09	Make announces a Make plugin listed in ChatGPT's plugin directory	community.make.com/t/feature-spotlight-make-plugin-for-chatg
2026-03-24	Make announces it is a built-in connector in Claude, set up with OAuth and no URL or token	help.make.com/make-now-a-built-in-connector-on-anthropic-cla
2026-01	Make connector listed in the Claude directory (Anthropic verified), remote URL https://mcp.make.com	claude.com/connectors/make

FAQ

Is this official?

Yes. Make hosts and documents the server on its developer site, and the Claude directory lists Make's connector as Anthropic verified.

Is it free?

Make says scenario-run tools work on all plans and management tools need a paid plan. Make's MCP docs don't say how agent-triggered runs count toward plan usage.

Does it change data in Make?

Yes. Management tools create, update and delete scenarios, data stores and records, and scenario runs act immediately in the apps each scenario connects to.

Is my data used to train models?

Make's MCP documentation doesn't say. Check Make's privacy policy and your AI provider's terms for how prompts and tool results are handled.

Can my accountant use it on my behalf?

Make doesn't document delegation. Access follows the account that signs in or owns the token, so an accountant would connect with their own Make user in your organization.

Something wrong or out of date? Tell us and we'll email you when it's fixed.

Sources

Verified by theFinanceStack, 21 Oct 2026. Vendor list on the CFO Software Market Map courtesy of Jason Staats; AI connectivity data by theFinanceStack. Not tax, compliance or financial advice.