Zapier MCP Pricing: Fix It Before Dec. 1 (2026)

Matt Payne··Updated ·8 min read
Key Takeaway

Zapier MCP costs 2 tasks per successful call. Dec 1 billing hits Next Gen Zaps. Audit your calls, move fixed work into Zaps, route only unclear records to AI, and cache repeated data.

Zapier MCP Pricing: Fix It Before Dec. 1

Zapier made ChatGPT look like a sales agent with 40,000 possible actions.

That access is useful, but every tool call adds to your bill.

Starting December 1, 2026, Next Gen Zaps will draw from your Zapier task pool. An agent making eight successful MCP calls per lead could burn 8,000 tasks across 500 leads.

That doesn't include model costs, retries, or duplicate research.

Step 1: Count Actions, Not "Agent Runs"

Zapier MCP is a hosted bridge between an AI client and Zapier. ChatGPT, Claude, and Cursor can use it to take actions in connected apps.

Zapier markets access to more than 9,000 apps. Its pages list between 30,000 and 40,000 possible actions.

The key number is two.

A Cursor HQ report on Zapier MCP billing says each successful MCP call uses two tasks. Failed calls don't count.

That means:

PlanIncluded tasksApproximate successful MCP calls
Free10050
Professional750375
Team2,0001,000

Professional was listed at $19.99 per month, billed annually. Team was listed at $69 per month. Larger task pools require higher tiers or overage billing.

Don't count conversations. Count successful actions.

"Research this lead and add it to HubSpot" might trigger:

1. Search HubSpot for the contact. 2. Search HubSpot for the company. 3. Fetch enrichment data. 4. Read the company website. 5. Create or update the contact. 6. Create or update the company. 7. Add the contact to a sequence. 8. Send a Slack notification.

That's eight successful MCP tool calls.

At two tasks each, one lead costs 16 tasks. Five hundred leads cost 8,000 tasks.

Start with an action audit.

Export Zap History. Record the trigger, each successful action, every AI step, and every retry.

Use this formula:

> Monthly tasks = runs × [(MCP calls × 2) + billed app actions + AI-step tasks] + retry buffer

Zapier says AI steps can cost one, three, or five tasks. The multiplier depends on the model tier you select.

Use a 15% retry buffer until your own history shows a better number.

Step 2: Build a Real Zapier MCP Cost Model

The December 1 change has two parts.

Direct MCP tool calls use two tasks per successful call. Next Gen Zap app actions and external calls will also draw from the shared task pool.

Zapier's billing guide for Next Gen Zaps and the Zapier SDK says each successful app action or external call costs one task. Filters, formatting, loops, and branches remain free.

Zapier's AI pricing adds another layer.

According to Zapier's AI by Zapier announcement, an AI step may use one, three, or five tasks. Zap History shows the model and task count.

Here's a working calculator:

InputExample
Monthly records500
MCP calls per record8
Tasks per MCP call2
AI steps per record1
AI task multiplier3
Retry buffer15%
Total10,925 tasks

The math is:

`500 × [(8 × 2) + 3] × 1.15 = 10,925`

Don't turn that into dollars with a random task price.

Your dollar cost depends on your plan, included tasks, and overage rate. Use this formula instead:

> Monthly cost = plan fee + overage tasks × contracted overage rate

Before December 1, qualifying Next Gen Zaps were free during early access. After that date, successful actions use the same task pool.

Your workflow didn't change. The billing did.

Easy connections can create unpredictable bills when nobody tracks volume.

Step 3: Put Fixed Work in a Deterministic Core

AI shouldn't decide things your workflow already knows.

If every qualified lead goes into HubSpot, make that a fixed action. If every new meeting alerts Slack, make that fixed too.

Don't ask ChatGPT to rediscover those rules 500 times.

Build a deterministic core with Zapier filters, paths, formatting, storage, and fixed actions. Zapier lists several internal utility steps as free, including Filter, Formatter, Path, Delay, Looping, Sub-Zap, Digest, and Storage.

A "By Zapier" label doesn't make every action free.

A Gmail send, HubSpot update, or external API request can still use tasks. The exemption applies to named utility steps.

A better AI BDR workflow looks like this:

1. Receive the lead through a webhook. 2. Normalize the email and company domain. 3. Check a stored deduplication key. 4. Reject personal emails with a fixed rule. 5. Read existing CRM fields. 6. Send unclear records to AI. 7. Route the result with Paths. 8. Write only approved changes. 9. Require approval before outbound messages.

The model decides whether the lead fits.

The workflow handles storage, routing, field mapping, and notifications.

Agents don't track task use on their own. Give one eight tools and it may use six when two would work.

Zapier's October announcement recommends mixing agentic and deterministic steps for better cost control and more predictable output.

Use AI for decisions. Use the workflow for plumbing.

Step 4: Route Only Uncertain Cases to AI

Decision-layer routing means AI handles only records that fixed rules can't resolve.

Start with three routes:

RouteRuleAction
Automatic passClear matchContinue without AI
AI reviewMissing or conflicting dataRun one AI step
Human reviewHigh-risk or low-confidencePause for approval

Suppose your AI BDR handles 500 leads each month.

A bad workflow sends all 500 through a three-task AI step. That uses 1,500 tasks before any CRM actions.

A routed workflow might resolve 70% with fixed rules. Only 150 leads reach AI.

That cuts AI use from 1,500 tasks to 450.

The 70% rate is an example assumption. Replace it with your Zap History data.

Use routing rules like these:

```text IF existing_customer = true THEN stop outreach

IF employee_count < 10 THEN route to nurture

IF industry AND country AND employee_count are present THEN score with fixed rules

IF two required fields are missing THEN send to AI research

IF confidence < 0.80 THEN require human approval ```

Content operations need the same restraint.

Don't ask AI whether every approved article should publish to WordPress. The editorial calendar already answered that.

Use AI for unclear work:

  • Pick the strongest angle.
  • Classify the audience.
  • Compare claims against source material.
  • Flag unsupported statements.
  • Choose between two campaign paths.

Use fixed actions for publishing, tagging, scheduling, and notifications.

Generic outbound damages trust at scale. An AI BDR should make fewer, better decisions.

Step 5: Add Caching, Batching, and Hard Caps

Caching stops your workflow from paying for the same answer twice.

Store company research by domain. Save the source, result, and expiration date.

A basic cache record needs five fields:

FieldExample
Key`acme.com`
Data type`company_research`
ResultStructured JSON
Source date`2026-11-20`
Expires`2026-12-20`

Check the cache before calling a research tool.

Use different expiration windows for different data:

  • Company description: 30 days
  • Employee count: 14 days
  • Funding news: 24 hours
  • CRM ownership: five minutes
  • Published content URL: no expiration

Batching helps when the connected app supports bulk actions.

Putting 100 records in a loop doesn't create one task. If Zapier performs 100 successful external actions, expect 100 billed actions.

Batch when the API accepts lists. Rate-limit everything else.

Use this control template:

```text MAX_RECORDS_PER_RUN = 100 MAX_MCP_CALLS_PER_RECORD = 3 MAX_AI_TASKS_PER_RUN = 75 CACHE_CHECK_REQUIRED = true RETRY_LIMIT = 2 HUMAN_APPROVAL_REQUIRED_FOR = send_email, delete_record, change_deal_stage ```

Zapier says AI runs pause for human review after crossing a 75-task cap. The company says that limit may become adjustable.

Three example budgets show the impact:

WorkflowNaive designRebuilt design
AI BDR, 500 leads8 MCP calls each = 8,000 tasks2 core calls each, 30% AI review, 20% outbound = about 2,850 tasks
Daily content, 30 days10 MCP calls daily plus AI = about 690 tasks4 MCP calls daily plus AI = about 330 tasks
CRM enrichment, 5,000 records4 MCP calls each = 40,000 tasksProcess only 20% changed records at 3 calls each = 6,000 tasks

These are planning examples, not Zapier quotes. Your task history determines the real budget.

Migration Checklist

  • [ ] Export 30 days of Zap History.
  • [ ] Count successful MCP calls by workflow.
  • [ ] Record each AI model multiplier.
  • [ ] Find loops containing external actions.
  • [ ] Remove duplicate CRM searches.
  • [ ] Add a cache before research calls.
  • [ ] Move fixed decisions into Filters and Paths.
  • [ ] Route only unclear records to AI.
  • [ ] Set retry limits.
  • [ ] Require approval for outbound actions.
  • [ ] Add a per-run task cap.
  • [ ] Compare monthly demand with your included tasks.
  • [ ] Test the rebuilt workflow with 50 real records.
  • [ ] Review cost per completed outcome after seven days.

StoryPros builds AI agents around measurable outcomes.

A sales agent is useful when it books meetings without wasting cash or trust.

FAQ

Is Zapier MCP free?

Zapier MCP is available through Zapier plans, but successful calls use tasks. Each successful MCP tool call uses two tasks, so the 100-task Free plan supports about 50 successful calls.

How much does Zapier MCP cost?

Zapier MCP doesn't have a separate listed fee. Usage comes from your Zapier task allowance. Professional starts at $19.99 per month for 750 tasks, and Team starts at $69 per month for 2,000 tasks when billed annually.

What is AI by Zapier?

AI by Zapier is an AI step inside a Zap. It can reason, choose approved tools, and take actions. Paid model tiers use one, three, or five tasks per step.

How can I connect ChatGPT to Zapier?

Create a Zapier MCP server, connect approved apps, and add that server to an MCP-compatible ChatGPT client. Limit permissions to the actions ChatGPT actually needs.

How do I reduce Zapier MCP pricing?

Cut successful tool calls with fixed Zaps, decision-layer routing, caching, and bulk actions. Set task caps and retry limits before you run AI BDR workflows at volume.

Related Reading

AI Answer

How much does each Zapier MCP tool call cost in tasks?

Each successful Zapier MCP tool call uses two tasks. A workflow making eight MCP calls per lead burns 16 tasks per lead. At 500 leads per month, that totals 8,000 tasks before AI steps or retries are counted.

AI Answer

What happens to Zapier MCP billing on December 1, 2026?

Starting December 1, 2026, Next Gen Zaps draw from your shared Zapier task pool. Each successful app action or external call costs one task. AI steps cost one, three, or five tasks depending on the model tier selected.

AI Answer

How can I cut my Zapier MCP task usage?

Move fixed decisions into Zapier Filters and Paths, which are free. Route only unclear records to AI. In one example, routing 70% of leads through fixed rules cut AI task use from 1,500 tasks to 450 tasks per month.