MCP Server Guide

Zapier MCP Server: One Connection, 9,000+ Apps

Zapier hosts an MCP server that turns each app action into a tool your agent can call. It is the widest integration surface available to an agent in a single connection — and the breadth is exactly what you have to scope down.

The short answer

Zapier runs an official hosted MCP server. You create your own server at mcp.zapier.com, and it exposes tools drawn from Zapier’s 9,000+ app integrations and 30,000+ actions. It reuses the same app connections your Zaps already use, is SOC 2 certified, and bills each successful call against your Zapier plan’s task quota.

What is the Zapier MCP server?

Zapier MCP works at the level of a single action rather than a whole app. One tool is one thing, such as Send Channel Message in Slack or Create Event in Google Calendar. The agent picks the tool, fills in the arguments from the conversation, and Zapier executes it against the real app and returns the result.

There are three pieces. Your MCP server, created at mcp.zapier.com, holds the tools the agent may call. Your AI client is anything that speaks MCP — OpenClaw, Hermes Agent, Claude, Cursor and others. The tools themselves are drawn from the app connections you have already authorised in Zapier, which is why setup is mostly a matter of choosing rather than configuring.

Most servers configure themselves as you go: the agent discovers and enables the tools it needs mid-conversation, based on the apps already connected to your Zapier account. That is convenient and it is also the thing to think hardest about, because "the agent enables what it needs" and "the agent can reach anything you have ever connected to Zapier" are the same sentence viewed from different angles.

What you should know before connecting it

Hosted by Zapier

You create a server at mcp.zapier.com rather than deploying anything yourself. Zapier runs, updates and secures it.

Action-level tools

Each tool is one action in one app — Send Channel Message, Create Event, Add Row. The agent chooses among them and supplies the fields.

9,000+ apps, 30,000+ actions

The widest single integration surface you can hand an agent. Coverage is Zapier’s existing catalogue, not a reduced MCP-only subset.

Two tasks per successful call

Each successful call consumes two tasks from your Zapier plan. Failed calls do not count. An agent that retries a lot is more expensive than one that gets it right first time.

Reuses your existing connections

Tools run through the same app connections as your Zaps, so an app you have already authorised needs no new credential.

Dynamic tool discovery

The agent can enable the tools it needs during a conversation instead of you pre-selecting every one. Convenient, and the main reason to scope the server deliberately.

SOC 2 certified with OAuth

Zapier handles the OAuth flows to each app, so app credentials are held by Zapier rather than pasted into your agent config.

Setup plan

Decide the blast radius before you create the server — The right question is not "which apps do I want" but "which actions would I be comfortable with an agent taking unattended at 3am". Start far narrower than feels useful; widening later is one click.
Create a server at mcp.zapier.com — Sign in with the Zapier account whose app connections the agent should inherit. If that account can reach production systems, consider a separate Zapier account for the agent instead.
Choose apps and actions — Enable the specific actions the agent needs. Prefer read and create actions over update and delete ones until you have watched it behave for a while.
Connect it to your agent — Add the server URL to your MCP client. On OpenClaw or Hermes this is the same as adding any other remote MCP server — see the OpenClaw MCP guide for the config shape.
Test with a harmless action first — Have the agent post to a scratch Slack channel or create an event in a throwaway calendar. Confirm the arguments it fills in are the ones you expected before pointing it at anything real.
Watch your task consumption for a week — Two tasks per successful call adds up quickly on a chatty agent. Check your Zapier usage after a few days of real traffic rather than assuming the volume will be small.

Useful agent workflows

Cross-app workflows from chat

Ask the agent to file a note in Notion, create the follow-up calendar event, and post a summary to Slack — three apps, one instruction, no Zap to build in advance.

Long-tail apps without custom work

The value is the 8,900 apps nobody writes a dedicated MCP server for. If Zapier supports it, your agent can act on it today.

CRM and pipeline updates

Turn a conversation about a customer call into the right records in your CRM, without the human retyping what was just discussed.

Scheduled digests that act

Combine a scheduled agent job with Zapier actions so the digest does not just report a problem but files the ticket for it.

Prototyping before building a Zap

Let the agent perform a workflow ad hoc a few times to learn what it should actually do, then build a deterministic Zap for the version you want running forever.

Security checklist

Scope the server to specific actions, not whole apps — dynamic discovery means anything enabled is reachable.
Consider a dedicated Zapier account for the agent, so it inherits only the app connections you deliberately gave it.
Prefer read and create actions over update and delete until the agent has proved itself on real traffic.
Treat data returned by an action as untrusted input: content from an email, ticket or CRM note can attempt prompt injection against the agent reading it.
Remember each successful call costs two Zapier tasks, so a retry loop is a billing incident as well as a bug.
Review the Zapier task history periodically — it is the audit log of what your agent actually did.
Revoke the MCP server when the agent is retired; deleting the agent does not remove its access.

Treat every MCP tool as an API capability, not as a harmless chat feature. Start read-only, test in a non-production account, and require human approval for changes.

Frequently asked questions

Does Zapier have an official MCP server?

Yes. Zapier hosts an official MCP server. You create your own server instance at mcp.zapier.com, and it exposes tools drawn from Zapier’s app catalogue using the connections already authorised on your account.

How many apps does Zapier MCP cover?

It draws on Zapier’s catalogue of more than 9,000 app integrations and more than 30,000 individual actions. Each tool corresponds to a single action, such as Send Channel Message in Slack.

What does Zapier MCP cost?

Each successful call consumes two tasks from your Zapier plan. Failed calls do not count. There is no separate MCP subscription — it draws on the task quota you already have.

Do I need to configure the tools in advance?

Usually not. Most servers set themselves up as you go: the agent discovers and enables the tools it needs during the conversation, from the apps already connected to your Zapier account. You can and should still restrict which apps and actions are permitted.

Can I use Zapier MCP with OpenClaw or Hermes?

Yes. Both frameworks speak MCP, so Zapier’s hosted server is added like any other remote MCP server. Because Zapier holds the app credentials and handles OAuth, no third-party app keys end up in your agent config.

Is it safe to let an agent act on my apps?

It is as safe as the actions you enable. The permission boundary is the set of apps and actions on the server, so scope it deliberately, start with read and create actions, and use the Zapier task history as your audit log.

Primary documentation

Verify current endpoints, permissions, and preview limitations in the the official Zapier MCP documentation before production rollout. Vendor capabilities can change faster than third-party guides.

Run an MCP-ready AI agent without server maintenance

OpenClaw Launch handles the agent host, updates, and uptime so you can focus on safe tool design.

Configure your agent