Zapier MCP Review (2026): 30,000+ Actions, One Server Endpoint
Zapier MCP gives your agent access to 30,000+ actions across 8,000+ apps without building individual integrations. Cloud-hosted, OAuth-authenticated, and zero local setup required. The widest automation coverage available through a single MCP endpoint.
Zapier MCP is the fastest path to broad SaaS coverage in an agent. You get a single cloud-hosted endpoint that exposes 30,000+ pre-built actions across 8,000+ apps - no local install, no per-service OAuth dance in your own code, no API key juggling. The trade-off is real: you are taking on Zapier's uptime as a dependency, latency is noticeably higher than local servers, and the action abstraction occasionally smooths over details you would handle differently against a native API. For agents that need to touch many different SaaS tools, this is the right call. For anything where you need raw API control or sub-100ms response times, reach for a native integration.
Pros and Cons
What works
- 30,000+ actions across 8,000+ apps - the widest coverage of any MCP server
- Zero local install - cloud-hosted, connects via HTTP/SSE
- Zapier handles the OAuth credential layer for every integration
- Works with any MCP client out of the box (Claude Desktop, Cursor, custom agents)
- Conditional logic, filters, and multi-step workflows via existing Zapier infrastructure
- Pre-configure which actions are exposed - keeps the tool surface manageable
- Setup time measured in minutes, not hours
What doesn't
- Latency is higher than local servers - HTTP/SSE round trips add up
- Zapier uptime becomes your agent's uptime - one SLA, no fallback
- Free tier is too restricted for real agent workloads
- Action abstractions can lose fidelity vs native APIs for complex calls
- Rate limits are Zapier's, not yours - shared infrastructure ceilings
- Quirks appear in some actions that work fine via direct API but behave unexpectedly through Zapier's layer
Scored Criteria Breakdown
Each criterion scored out of 10 against our standard methodology.
Latency is the one score that drags the overall down. Cloud-hosted MCP with OAuth token exchange on each call adds overhead that local servers simply don't have. For most automation use cases this doesn't matter. For tight agentic loops where the model is calling tools rapidly, it is worth knowing.
When to Use It / When Not To
Good fit
- Agents that need to reach many different SaaS tools without per-integration plumbing
- Automation workflows: CRM updates, ticket creation, calendar scheduling, email sends
- Prototyping and early-stage agents where breadth matters more than precision
- Teams already on Zapier who want to expose existing Zaps to an agent
- Non-technical teams building agents via Claude Desktop where local installs are impractical
- Multi-step orchestration using Zapier's filters and conditional logic
Not a good fit
- Latency-sensitive loops where every tool call adds visible delay
- Workloads that hit Zapier's rate limits at scale - you can't negotiate those ceilings
- Cases where you need full API fidelity and Zapier's action abstraction loses detail
- High-volume production agents where Zapier's pricing becomes a cost problem
- Offline or air-gapped environments - cloud-only, full stop
- Anything requiring custom auth flows not covered by Zapier's OAuth patterns
What It's Actually Like to Use
The action selection UX is where Zapier MCP earns its score. Before you connect the server to any client, you go into zapier.com/mcp and pick exactly which actions you want to expose. This matters more than it sounds: without curation, a 30,000-action surface would be unusable for any LLM trying to reason about which tool to call. The pre-configuration step forces you to scope it down to 10-30 relevant actions, which is a good constraint.
Auth is the other standout. Every integration authenticates through Zapier's existing OAuth layer, which means your agent never handles raw credentials. The token refresh cycle is Zapier's problem. For multi-integration agents - say, something that reads from Google Sheets, writes to HubSpot, and sends a Slack message - this removes a substantial amount of credential management code from your stack.
The latency reality: in testing, simple actions (create a HubSpot contact, add a row to Sheets) came back in 800ms-2s depending on load. That's fine for task automation where a human is waiting on a result. It's a compounding problem in a tight agentic loop where the model is calling five tools in sequence - you're adding 4-10 seconds to the total time. Know your workload before committing.
One genuinely useful feature: because each action maps to a Zapier step, you can chain it with existing Zap logic. If you already have a Zap that does post-processing on a HubSpot contact (territory assignment, tag normalization, whatever), an agent writing via Zapier MCP inherits that logic automatically. You're not bypassing your existing automation layer - you're exposing it.
Pricing Reality Check
The free tier gets you connected and lets you test the flow, but task limits kick in fast for any real agent. Budget Starter as the baseline. If you're running agents that trigger dozens of Zapier actions per day, do the math against Professional before you build - unexpected overage costs are a common frustration in production.
How to Get Started
No local install, no npm package. This is the whole flow:
Go to zapier.com/mcp and configure your actions
Log in to Zapier, navigate to the MCP section, and select which actions you want to expose. Connect the relevant integrations (Google, Slack, HubSpot, etc.) via OAuth. This pre-configuration step determines what your agent can call - keep it scoped.
Copy your personal MCP server URL
Zapier gives you a unique HTTP/SSE endpoint URL. It's authenticated to your account and includes your selected action set. Treat this like an API key - don't commit it to public repos.
Add to your MCP client config
For Claude Desktop, add to claude_desktop_config.json:
{
"mcpServers": {
"zapier": {
"url": "https://mcp.zapier.com/api/resources/s/YOUR_ID/mcp",
"transport": "sse"
}
}
}For Cursor or any client that accepts SSE transport, the pattern is the same - paste the URL into the server config. Restart the client.
Test a call
In Claude Desktop, try asking it to perform one of the actions you configured. If Zapier's OAuth tokens are fresh, it works immediately. If you see auth errors, re-authorize the relevant integration in your Zapier account.
Total setup time for a developer already familiar with MCP client configs: under 10 minutes. The bottleneck is usually re-authorizing integrations in Zapier if you haven't used them recently.
Add Zapier MCP to Your Agent
30,000+ actions, zero local install, auth handled. If you're building agents that need broad SaaS coverage, this is the fastest starting point available.
Visit site Free tier available · Starter from ~$20/mo