Zapier MCP is a hosted Model Context Protocol server that lets any MCP-compatible AI — Claude, ChatGPT, Cursor, GitHub Copilot — take real-world actions across 9,000+ apps without custom API work. It launched in March 2025, runs on every Zapier plan including Free, and costs 2 tasks per tool call from your plan's shared quota.
For operators building AI agent workflows for clients, this is the fastest path from "I want my AI to update HubSpot when a call ends" to a working system. Here's what it does, what it actually costs to run, and the service angle.
What Zapier MCP does
Normal Zapier runs automated Zaps: a human designs a trigger-action flow ahead of time (new form submission → create CRM record → send Slack message). The AI follows the rails.
Zapier MCP works differently. Connect your AI client to the MCP server once, and the AI can take any of 30,000+ actions across 9,000+ apps on demand — no pre-built flow required. The AI decides what to do based on your prompt.
Example: you ask Claude, "Send a Slack message to #sales-team summarizing this call transcript, create a HubSpot deal for the contact, and add a Google Calendar follow-up for next Tuesday." Claude calls the Zapier MCP server three times. All three things happen. No Zap built.
Verified use cases from Zapier's own documentation:
- Email: draft and send Gmail messages, filter and reply to threads
- CRM: find, create, and update HubSpot and Salesforce records
- Calendar: create Google Calendar events with attendees and meeting agenda
- Team comms: post formatted Slack messages to specified channels
- Data logging: append rows to Google Sheets on command
- Content pipelines: pull a Google Doc draft → generate a social post → schedule in Buffer → notify in Slack — all in one prompt
- Post-call routing: AI voice agents log call outcomes, trigger confirmation emails, and route leads without a separate Zap
The underlying protocol (MCP) was created by Anthropic and is now standard across Claude, ChatGPT, Cursor, GitHub Copilot, and most modern AI coding tools.
What it costs
Zapier MCP is included on every plan — Free, Professional, Team, and Enterprise — at no additional charge. The cost model is task consumption: each MCP tool call uses 2 Zapier tasks from your plan's shared quota (same pool as Zaps and AI steps).
| Plan | Annual price | Tasks/month | MCP calls/month | |---|---|---|---| | Free | $0 | 100 | ~50 | | Professional | ~$19.99/mo | 750 | ~375 | | Team | ~$69/mo | 2,000 | ~1,000 | | Enterprise | Custom | Custom | Custom |
A 10-step agent workflow burns 20 tasks. On Free (100 tasks/mo), that's roughly 5 complete agent runs. For any real client deployment, the client pays for their own Zapier plan — put it on their card, not yours.
Zapier MCP vs. Make MCP
Make launched its own MCP integration in May 2025. The comparison matters because they work architecturally differently.
| | Zapier MCP | Make MCP | |---|---|---| | Apps | 9,000+ | 3,000+ | | Actions | 30,000+ | 30,000+ | | How the AI acts | Agentic — any action on demand | Calls pre-built Make scenarios | | Setup time | Fast — connect once, prompt freely | Slower — scenarios must be built first | | Cost at high volume | Can exceed $300/mo at 100k ops | Often under $100/mo at same volume | | Best for | Breadth, speed, non-technical operators | Cost control, complex logic, technical builders |
The core tradeoff: Zapier MCP is agentic; Make MCP is structured. Zapier lets the AI act freely across your full app library. Make gives the AI access only to scenarios a human already built — more control, lower running cost, more upfront work.
For most operators selling to SMB clients, Zapier wins on speed: you can demo a working system in a client account within an hour. The cost gap matters once you're running high volumes.
The service you sell with it
Zapier MCP's real value for operators is deploying it inside client stacks, not just your own.
The pitch: AI agents that connect directly to the client's existing apps. No separate AI subscription. No new tool to learn. Their AI just starts doing things in HubSpot, Gmail, and Google Calendar.
The flagship build: post-call automation. After every sales or service call, an AI logs the summary to the CRM, sends a follow-up email from the rep's inbox, and books the next touch. Runs entirely through Zapier MCP. No custom code. 1-2 days to build once you have a template.
What to charge:
- Setup: $1,500-$2,500 — MCP configuration, permission scoping, testing, documentation
- Retainer: $300-$600/mo — monitor task usage, add new actions, keep connections healthy
Pair it with a broader Zapier automation audit and the upsell surface expands significantly: every manual workflow you find in discovery is a potential MCP-powered AI action. For how the retainer model pencils out, see how to start an AI automation agency — same economics, same packaging.
How to connect your AI client
The technical setup takes under 15 minutes:
- Go to zapier.com/mcp and generate your MCP server
- Choose which apps and actions to expose (you control exactly what the AI can touch)
- Copy the MCP server URL
- Paste it into your AI client's MCP settings (Claude Desktop, Claude Code, Cursor, ChatGPT, etc.)
The permission scoping step matters for client deployments: restrict the AI to specific apps first (Gmail but not Salesforce, calendar read but not create), then expand as trust builds. A scoped MCP server is also easier to hand off to clients — they can see exactly what the AI is allowed to do.
Works with: Claude Desktop, Claude Code, ChatGPT, Cursor, GitHub Copilot, Windsurf, and any client that implements the open MCP standard.
FAQ
Does Zapier MCP work with Claude?
Yes. Claude Desktop and Claude Code both support MCP natively. Paste the Zapier MCP server URL into Claude's MCP settings and the AI can immediately take actions across any connected app. Claude handles the reasoning and task decomposition; Zapier handles execution.
Is Zapier MCP free?
MCP access is included on every Zapier plan, including Free. There's no separate MCP line item — you pay through task consumption. Each MCP tool call uses 2 Zapier tasks. The Free plan includes 100 tasks/month, giving you roughly 50 tool calls.
How is Zapier MCP different from regular Zaps?
Zaps are pre-built, trigger-action workflows a human designs ahead of time. MCP is agentic — the AI decides what actions to take based on a prompt, on demand, without a pre-built flow. Think of Zaps as "if this, then that" and MCP as "AI agent with hands."
What's the difference between Zapier MCP and Make MCP?
Zapier MCP gives an AI free-range access to 9,000+ apps on demand, with no pre-built scenarios required. Make MCP requires the AI to call Make scenarios a human already built — more control and lower cost at scale, but more setup time. See the comparison table above for when each makes sense.
Can I use Zapier MCP for client work?
Yes — this is the highest-value use case for operators. Set up MCP inside the client's Zapier account, scope it to the apps they already use, and you have an AI agent living inside their existing stack. No new software purchase required on their end. Charge $1,500-$2,500 for setup, $300-$600/mo to maintain.