Does the Zapier MCP server work with Claude?

Yes. Zapier MCP is a remote MCP server that works with any MCP client, including Claude Cowork, Claude Code, and Claude Desktop. You generate a server URL and connect it; Claude can then call the Zapier actions you've enabled.

Is there an official Zapier MCP server?

Yes. Zapier builds and hosts its own MCP server at mcp.zapier.com. It is a managed, remote server — there's no separate community server to install, because the value is Zapier's own app directory.

Is the Zapier MCP server free?

MCP is available on all Zapier plans, including the free tier. But each MCP tool call uses two tasks from your plan's quota — the same task bucket your Zaps use — so heavy agent usage burns through tasks quickly.

How many apps and actions does Zapier MCP support?

Zapier MCP exposes 8,000+ connected apps and 30,000+ pre-built actions. You choose which specific actions to expose to your AI per server, so the assistant only sees the tools you enable.

Zapier MCP Server: What It Does & How to Use It with Claude

The Zapier MCP server lets an AI assistant take real actions across 8,000+ apps (sending emails, creating tasks, updating records) without you wiring up each API. Instead of exposing one product, it turns Zapier’s entire action library into MCP tools you can selectively enable for Claude.

What can the Zapier MCP server do?

Zapier MCP is a remote, hosted server: you don’t run anything locally. You build a server at mcp.zapier.com, add the specific actions you want, and Claude calls them.

Capability	What it enables
Run actions	Trigger any of 30,000+ pre-built actions (send a Slack message, create a Google Calendar event, add a HubSpot contact) from natural language
Search / read	Query connected apps for data (find a record, look up a contact) before deciding what to do next
8,000+ app catalog	Reach apps that have no official MCP server of their own by routing through Zapier’s existing integrations
Per-action scoping	Enable only the specific actions you want exposed; the assistant never sees the full catalog by default
Managed auth	Zapier holds the app credentials and handles the OAuth connection, so app tokens never reach the model
Action history	Every call is logged in your Zapier action history for audit and debugging

Which Zapier MCP server is official?

Zapier MCP is official and first-party. There is no meaningful community alternative, because the whole point is access to Zapier’s own integration directory.

Official Zapier MCP (mcp.zapier.com) is the only real option. It’s a remote server Zapier hosts and maintains, with two auth modes: an API key (best for personal use and local clients) and OAuth (best when end users connect their own Zapier accounts inside an app you’re building).
Community servers exist for individual underlying tools (e.g. a dedicated Slack or GitHub MCP server), but those replace Zapier, they don’t extend it. Use Zapier MCP when you want breadth across many apps from one endpoint.

How do you connect it to Claude (Cowork / Code / Desktop)?

Go to mcp.zapier.com and create a new MCP server. Pick your client (Claude is a built-in option).
Click + Add tool and add only the actions you need. Start with one or two read or search actions, not write actions.
Copy the generated server URL (with API key) or set up OAuth.
Add it to Claude: in Claude Desktop / Cowork, add it under Settings → Connectors as a remote MCP server using the URL; in Claude Code, run claude mcp add --transport http zapier <your-server-url>.
Confirm the tools appear, then test with a read-only action before enabling anything that writes or sends.

How we use it: Zapier MCP is great for reach, connecting Claude to a long tail of apps (calendars, CRMs, niche SaaS) where building a dedicated server isn’t worth it, and for quick “do this one thing in app X” prompts. It’s bad for high-volume agent loops: every tool call costs two tasks, so a chatty agent can drain a month of quota fast, and you’ll often get better latency and richer schemas from a tool’s own native MCP server when one exists. The realistic gotcha: each enabled action shows up as a separate tool, so adding 30 actions floods Claude’s tool list and the model starts picking the wrong one. Keep each server lean and purpose-built.

How do you keep it secure?

Scope it tightly. Enable only the specific actions a given server needs, start with read/search actions before write actions, and lean on Zapier’s managed auth so app credentials never reach the model. Audit usage in your Zapier action history, and use OAuth (not a shared API key) whenever real users are involved.

Wiring Zapier into your stack

The strength of Zapier MCP is breadth, but breadth is also where agents go wrong, picking the wrong tool or burning tasks on loops. We help teams decide where Zapier MCP is the right shortcut and where a purpose-built MCP server earns its keep, then wire the whole thing into Claude cleanly. Book a discovery call.

Sources: Zapier MCP docs · Zapier MCP usage & billing · Zapier MCP guide