Does Zapier have an MCP server, and do Make and n8n?

Does Zapier have an MCP server?

Yes. Zapier MCP is Zapier's own hosted MCP server. It lets Claude, ChatGPT, Cursor and other MCP apps run actions in more than 9,000 connected apps, signs in with OAuth, needs Streamable HTTP, and uses two tasks from your Zapier plan per successful tool call.

Make and n8n have official MCP servers too, and they work differently. This page compares all three from their own documentation, read October 30, 2026. If none of them fits the job, our MCP server development work covers custom servers built on your own accounts.

Zapier's MCP documentation describes it as "the Model Context Protocol interface to Zapier", with "9,000+ apps and 40,000+ actions". Zapier handles the sign-in to each connected app, and in its words "handles OAuth, credentials, and rate limiting". A server runs in one of two modes: in agentic mode the AI finds and adds the actions it needs during the conversation, and in managed mode you pick a fixed set of actions in advance.

Two details catch people out. Zapier MCP "does not support Server-Sent Events (SSE)", so an older client that only speaks SSE will not connect. And sharing a live server with colleagues "needs a Team or Enterprise plan"; on any plan you can share a copy of your tool list as a tool bundle instead.

How much does Zapier MCP cost?

There is no separate MCP fee. Zapier MCP runs on your existing plan, and each successful tool call uses two tasks. Failed calls and setup use none. On the plan prices we read, that is 50 calls a month on Free and 375 on Pro.

Zapier's MCP usage page says "Each successful tool call through your MCP server consumes two tasks" and "Tool calls that fail do not consume tasks." Its task usage help article, updated October 28, 2026, gives the same rate. MCP calls draw on the same task allowance as your Zaps. When the allowance runs out, calls stop until the cycle resets unless you upgrade or turn on pay-per-task billing.

We worked out what each plan's included tasks buy if nothing else on the account uses tasks. Plan figures are from Zapier's pricing page, read October 30, 2026, at each plan's entry tier.

In practice your Zaps share that allowance, so the real number of calls is lower. A team of five people each making ten tool calls on a working day (about 22 days) makes 1,100 calls, which is 2,200 tasks a month. That already passes the Team plan's entry tier before any Zap runs. For a full cost comparison of the two platforms, see our n8n and Zapier pricing research.

Does Make have an MCP server?

Yes. Make's MCP server is hosted by Make and runs over stateless Streamable HTTP. It can run your scenarios on any plan, and on paid plans it can also manage scenarios, connections, webhooks and data stores. You sign in with OAuth or an API key.

Make's MCP server documentation says it "allows AI systems, such as large language models (LLMs), to run scenarios and manage your Make account", and that "Scenario run tools are available to all plans, and management tools are available to paid plans." The addresses are:

• OAuth: https://mcp.make.com/v2
• API key in a header: https://<MAKE_ZONE>/mcp/v2 with Authorization: Bearer <API_KEY>
• API key in the address: https://<MAKE_ZONE>/mcp/u/<API_KEY>/v2, which puts the key in any log that records the address, so we avoid it

Timeouts matter for longer scenarios. A scenario run tool waits 25 seconds with OAuth or 40 seconds with an API key, and management tools wait 30 or 60 seconds. After a timeout, "the scenario continues running in Make for up to 40 minutes", and the AI can fetch the result later if you granted the scenarios:read scope. The page does not say how many credits an MCP call uses. If you build on Make, our Make automation service covers scenario design.

Does n8n have an MCP server?

Yes, two kinds. An instance-level MCP server gives one connection to your whole n8n instance, with OAuth or an API key. The MCP Server Trigger node turns a single workflow into its own MCP server. Both run on your n8n instance, Cloud or self-hosted.

n8n's instance-level MCP docs say an owner or admin turns on "Enable MCP access" in Settings, and the server address ends in /mcp-server/http. Nothing is exposed by default: you enable workflows one by one, or by project or folder. Connected AI apps can search and run enabled workflows, and from n8n 2.13.0 they can create and edit workflows and data tables. One limit to plan around: access is not scoped per client, so every connected AI app sees every MCP-enabled workflow the user can access.

The MCP Server Trigger node supports SSE and Streamable HTTP, with None, Bearer or Header authentication, and "doesn't support standard input/output (stdio) transport." Its production address only exists once the workflow is published. Our n8n automation team builds and hosts both kinds.

How do Zapier, Make and n8n MCP servers compare?

Zapier reaches the most apps and bills per call in tasks. Make runs and manages your scenarios, with management tools on paid plans. n8n runs on your own instance, can build workflows, and its docs state no per-call charge. All three sign in with OAuth.

We built this matrix from each vendor's documentation, read October 30, 2026. "Not stated" means the page we read did not say.

Which one should you use?

Use the one that already runs your automations, because the MCP server exposes what you have built there. Zapier suits teams that want many apps with little setup. Make suits scenario-heavy teams. n8n suits teams that want control and self-hosting.

The bigger question is whether you need an MCP server at all. If the job is a fixed process, such as a new lead going into the CRM and getting a text, a normal workflow is cheaper and easier to test than asking an AI app to do it. MCP pays off when people want to ask an AI app open questions about live data, or tell it to act. We cover that choice in the difference between an MCP server and an API.

Two cases point to a custom server instead. First, when one tool must combine several systems with your own business rules, such as checking a CRM record and a booking calendar before writing. Second, when you need strict per-person limits that the platform servers do not give you, such as n8n's lack of per-client scoping. Our Zapier automation services page covers the Zapier side if you stay there.

Why is a Zapier, Make or n8n MCP connection not working?

The usual causes are a client that only speaks SSE, a plan that has run out of tasks, a scenario that outlasts Make's timeout, an n8n workflow that was never published or enabled, and a proxy that strips MCP headers.

If the trouble is an n8n AI agent that ignores its tools rather than a connection error, see why an n8n AI agent does not call its tools.

More questions about Zapier, Make and n8n MCP servers

Is Zapier MCP free?

Zapier lists MCP on every plan, including Free, but every successful tool call uses two tasks from the plan's allowance. The Free plan has 100 tasks a month on the pricing page we read on October 30, 2026, which is 50 tool calls if no Zaps run. Setup and sign-in do not use tasks, and failed calls are not counted.

Can I use Zapier MCP with ChatGPT and Claude?

Yes. Zapier's docs name Claude, ChatGPT, Cursor, Grok Bot and VS Code among supported apps, and say any application that supports MCP can connect. The client must support Streamable HTTP. Most clients sign in with OAuth. Zapier also built a connector for Meta's Muse agent, which we covered in our Zapier and Muse news piece.

Does the Make MCP server work on the free plan?

Partly. Make's docs say scenario run tools are available on all plans, so an AI app can run your scenarios on Free. Management tools, which view and change scenarios, connections, webhooks and data stores, need a paid plan. The page we read does not say how many credits an MCP call uses, so test with a small volume first.

Is n8n's MCP server only for self-hosted n8n?

No. n8n's docs describe the same Instance-level MCP settings page for Cloud and self-hosted instances. Self-hosted owners can also switch the feature off entirely with the N8N_DISABLED_MODULES=mcp environment setting. On either, an owner or admin has to enable MCP access and choose which workflows to expose.

Can an AI app build Zaps, scenarios or workflows through MCP?

n8n's instance-level server can create and edit workflows from version 2.13.0. Make's management tools on paid plans can view and modify scenarios. Zapier's docs we read describe running actions, not building Zaps. Whichever you use, review any workflow an AI writes before you publish it, because it will run on real data.

Can you set one of these up for us?

Yes. We connect the right server to the AI apps your team uses, limit scopes and approvals, and test every tool. If a platform server cannot do the job, we build a custom one on your accounts. The strategy call is free; a small build of two or three workflows is about $1,000 at the time of writing, and larger work gets a fixed quote. Contact us to start.

Sources and further reading

• Zapier, Zapier MCP documentation, checked October 30, 2026.
• Zapier, MCP usage and task billing, checked October 30, 2026.
• Zapier Help Center, How is task usage measured in Zapier? (updated October 28, 2026), checked October 30, 2026.
• Zapier, Pricing, checked October 30, 2026.
• Make, Make MCP server, checked October 30, 2026.
• n8n, Accessing the n8n MCP server, checked October 30, 2026.
• n8n, MCP Server Trigger node, checked October 30, 2026.