Disclosure first, because it should change how you read this: we make BizOps Pilot, which competes with Zapier. We are not a neutral review site. What we can offer instead is that every number below comes from Zapier's own documentation, read on 16 July 2026 and independently re-confirmed against Zapier's live site on 22 July 2026, with the link and date next to it. The only thing we added is arithmetic, and we have marked every place we did it.
We are also going to tell you, in plain language, where Zapier is better than us. That section is not a formality. If you read to the end you will find at least one good reason to buy from them instead.
The rate
"Each successful tool call through your MCP server consumes two tasks. This is a fixed rate." docs.zapier.com/mcp/usage, read 16 July 2026
Three things travel with that rate, and all three are in Zapier's own docs (the same page, and help.zapier.com's MCP usage article, dated 4 June 2026):
- Failed calls do not count. Only successful ones.
- There is no separate MCP bill. MCP tool calls draw from the same monthly task allowance as your Zaps.
- There is no limit on the number of tool calls per session.
Zapier's MCP product page states the rate too, in a single FAQ line: "Each MCP tool call uses two tasks from your plan's quota" (zapier.com/mcp, read 22 July 2026).
So the rate is published, it is fixed, and it is not hidden anywhere. Hold onto that, because it matters later.
The examples Zapier published, and the one nobody finished
The same docs page gives worked examples. These are Zapier's numbers, not ours:
- "Add 5 rows to a spreadsheet" = 5 tool calls (10 tasks)
- "Send emails to 3 people" = 3 tool calls (6 tasks)
- "Search and update 10 records" = 11 tool calls (22 tasks)
docs.zapier.com/mcp/usage, read 16 July 2026
Look hard at the third one. You typed one sentence. It cost eleven tool calls.
Eleven reads as one search plus ten updates. That is our reading of their number, not something they spell out, but no other composition makes sense. And here is the thing worth sitting with: the number eleven is not in the sentence you typed. Nothing about the phrase "search and update 10 records" announces that it will be eleven calls rather than two. The fan-out happened inside the model, after you hit enter, on the far side of a decision you did not make.
Extend their own pattern and it gets sharper. Fifty records, same shape, is fifty-one calls and 102 tasks. That is our extension of their example, not their published one, so treat it as arithmetic rather than documentation. But if it holds, one sentence costs more than Zapier's entire free monthly allowance, which is 100 tasks (zapier.com/pricing, read 22 July 2026).
What a real session costs
Nobody uses an AI assistant for one instruction and then closes the laptop. So let us build a session out of Zapier's own three examples, because then every input is theirs and only the addition is ours.
Call it a Monday morning admin pass:
| What you asked for | Tool calls | Tasks |
|---|---|---|
| "Search and update 10 records" | 11 | 22 |
| "Send emails to 3 people" | 3 | 6 |
| "Add 5 rows to a spreadsheet" | 5 | 10 |
| One session | 19 | 38 |
Three sentences. Nineteen tool calls. Thirty-eight tasks.
Now against the plans. Zapier's free plan includes 100 tasks a month (zapier.com/pricing, read 22 July 2026). Professional starts at $19.99/mo billed annually and includes 750 tasks a month (zapier.com/pricing, read 22 July 2026). Every figure in the last two columns is our division:
| Plan | Price | Tasks/mo | MCP tool calls/mo | Sessions like the one above |
|---|---|---|---|---|
| Free | $0 | 100 | 50 | 2 |
| Professional | from $19.99/mo, billed annually | 750 | 375 | 19 |
| Team | from $69/mo, billed annually | 2,000 | 1,000 | 52 |
Team's base allowance used to be unpublished. It is not anymore: 2,000 tasks a month at that price (zapier.com/pricing, read 22 July 2026), which is 1,000 MCP tool calls a month and 52 sessions like the one above. More room than Professional. Still a ceiling, and still one your agent picks the speed toward, not you.
Read the free row again. The free plan is fifty MCP tool calls a month. Not fifty requests. Fifty calls, which the examples above suggest is closer to three requests than fifty. Two of our sessions and the third one halts twelve calls in.
Professional is more interesting, and this is where we have to be fair. Nineteen sessions a month sounds fine until you count working days. A month has about twenty-one of them. Twenty-one sessions at 38 tasks is 798 tasks, against an allowance of 750. Do that one three-sentence session every working day and you do not finish the month. You run out somewhere in the last week.
The part where we refuse to make it sound worse than it is
Here is the number that will not help our argument, so we are putting it in bold. On Professional, $19.99 for 750 tasks is about $0.027 a task, which makes an MCP tool call about 5.3 cents. Our nineteen-call session costs about a dollar. (Our arithmetic on their published figures.)
A dollar for a morning's admin is not a rip-off. It is fine. If somebody tells you Zapier MCP is outrageously priced per call, they have not divided.
The problem is not the price per call. It is that the call count is not yours to predict.
You did not choose eleven. You chose a sentence. The model chose eleven, and it chose it after you committed, and per Zapier's own docs there is no limit on how many calls a session can make. A meter needs you to forecast usage in order to pick a tier. An agent's whole value proposition is that you do not specify the steps. Those two things are pulling in opposite directions, and no amount of good faith from either side resolves it. A task meter and a conversational agent are a bad fit, structurally, not because anyone behaved badly.
When you cross the line, Zapier's pricing page says you are "switched to pay-as-you-go unless you turn it off or move up to a higher tier" (zapier.com/pricing, read 22 July 2026). The same page notes "the higher the task tier, the lower the cost per task," so volume does get cheaper. We did not read the higher tier prices off the page ourselves, so we are not quoting them.
A related meter, which we are not going to multiply for you
Separately from MCP, Zapier repriced AI steps on 15 June 2026, and the same rates still stand today. Per their help centre (confirmed 22 July 2026): AI by Zapier steps are priced by model tier, Standard 1x, Advanced 3x, Premium 5x, tool calls add to the base cost at the same rate, new steps default to Advanced (3x), and you can "connect your own AI account (1x)" instead.
That last clause is worth a pause. The largest automation company in the world has documented, on its own help page, that bringing your own AI key costs one third of its default tier. We think that is the single most interesting sentence Zapier published in 2026, and it is on their site, not ours.
But note what we are not doing. AI by Zapier steps and Zapier MCP tool calls are two different products with two different meters. You will see people multiply the 2 tasks per MCP call by the 3x AI default and quote you a scary number. Zapier's docs do not say to do that, so we are not doing it. If you want the AI-key argument, it stands on its own without being inflated.
One more from that page, and it is to Zapier's credit: if a step reaches 75 tasks in a single run, the Zap pauses and asks you to approve before continuing. That is a circuit breaker. It exists because somebody at Zapier knows exactly what we described above, that an AI step can fan out past what the person who wrote it expected, and they shipped a guard for it. That is good engineering and honest product design. It is also, read another way, the vendor conceding the whole point of this article.
One more thing worth naming plainly, because it changes the shape of the whole picture. Zapier now sells two more metered products alongside Zap workflows, AI by Zapier, and MCP: Agents, billed in "activities," and Chatbots, capped by how many chatbots your tier includes. Neither shares a meter with the other three (zapier.com/pricing, Agents and Chatbots tabs, read 22 July 2026). That is four separate consumption surfaces on one account now, and MCP tool calls are only one of them.
Where Zapier is genuinely better than us
Not a formality. Read this bit.
The catalogue, and it is not close. Zapier's MCP page advertises MCP access to 9,000+ apps (zapier.com/mcp, read 22 July 2026). We reach roughly 250 through Composio, plus native connectors, plus any HTTP API, plus any MCP server you point us at. That is more than an order of magnitude apart. If your stack has something regional, vertical, or obscure in it, Zapier has it and we very likely do not. The "big catalogues are mostly long tail" retort is true in aggregate and worthless if the app you need is in the tail.
Maturity. Error handling, replay, versioning, an app-review process, and a support organisation that exists. We are one person.
Their metering is more generous than people say. Triggers do not count, Filter and Paths steps do not count, and action steps that error or halt do not count. On MCP specifically, failed calls do not count (docs.zapier.com/mcp/usage, read 16 July 2026). You are billed for work that succeeded. That is a fair rule, and the affiliate posts claiming Zapier bills you for failures are simply wrong.
They published the damaging examples themselves. Everything in this article is arithmetic on numbers Zapier chose to print. A company hiding its costs does not put "search and update 10 records = 11 tool calls" in its own documentation. This is a documented design, not a scandal, and if we implied otherwise you should trust nothing else here.
Zapier is the safe institutional choice. If you need to justify the decision to somebody else, or you need the vendor to still exist in three years without thinking about it, that is a real requirement and Zapier answers it. We do not, yet.
What we do instead, and what it does not fix
We do not meter runs. That is the whole answer, and it makes this entire article inapplicable to us: there is no multiplication to do, because there is no number to multiply. BizOps Pilot is $39/mo flat for Pro, with unlimited runs on every plan including the free beta tier. The free tier bounds how many Pilots and Systems run at once (three and one), which is a capability ceiling, not a volume meter. An active Pilot on the free plan can run forever, as often as its trigger fires. There is also a Founding option, $249 once, hard-capped at 300 seats.
Four honest caveats, because a page like this is worth nothing if we spend the credibility we just built:
Our MCP server is a Pro feature. It is not on the free tier. If you came here for MCP specifically, our free plan does not give it to you, and Zapier's does give you fifty calls a month. That is a real point for them.
Removing our meter did not remove the fan-out. Your agent still made eleven calls. With us those calls cost you no tasks, but the tokens go on your own Anthropic or OpenAI bill, at their prices, with nothing added by us. That is the trade: you bring the key, you pay the provider directly, we do not count anything. The fan-out is still real, it just shows up somewhere we do not profit from. We removed our meter, not the physics.
Zapier's 75-task circuit breaker is a feature we admire and do not have. Under a bring-your-own-key model the runaway budget at risk is the customer's own provider bill, which is arguably worse. A guard is the right idea and they shipped it first.
We are pre-launch and small. No customer counts, no case studies, no testimonials, because we do not have any yet. One founder answers the email. And if you are technical, n8n self-hosted has no meter either, brings your own key with no credit limits, and costs the price of a VPS, which is less than $39. We beat it on there being no server to run and nobody to page at 3am. We do not beat it on price, and we are not going to pretend to.
The short version
- Zapier MCP costs 2 tasks per successful tool call, fixed, from your normal plan allowance. Failed calls are free. (docs.zapier.com/mcp/usage, 16 July 2026)
- That is 50 tool calls a month on the free plan and 375 on Professional at $19.99/mo billed annually with 750 tasks. (Our division of their figures.)
- A three-sentence session built from Zapier's own worked examples costs 38 tasks. Once every working day does not fit inside Professional's 750.
- Per call, that is about 5.3 cents, which is reasonable. The problem is not the price. The problem is that your agent picks the call count and you find out afterwards, and Zapier's docs confirm there is no per-session cap.
- All of this is published by Zapier, in public, with worked examples. It is a design decision, not a trick.
If you want to check us, do not take our word for it. Open docs.zapier.com/mcp/usage, read the three examples, and do the addition yourself. It takes a minute, and it is the only part of this article that is ours.
Sources
Most figures below were first read on 16 July 2026 and independently re-confirmed against Zapier's live site on 22 July 2026. The MCP worked examples and the 4 June 2026 help centre article were read on 16 July 2026 and not re-fetched this pass, though nothing found on 22 July contradicts them. Prices and rates in this category change without notice, including ours. Every number here is a dated snapshot, not a promise.
- Zapier MCP usage, 2 tasks per successful tool call, worked examples, no per-session cap: https://docs.zapier.com/mcp/usage (read 16 July 2026)
- Zapier help centre, how Zapier MCP usage works (dated 4 June 2026): https://help.zapier.com/hc/en-us/articles/45645738385805-How-Zapier-MCP-usage-works (read 16 July 2026)
- Zapier MCP product page, "9,000+ apps", FAQ line on 2 tasks per call: https://zapier.com/mcp (read 22 July 2026)
- Zapier's task-rate page, the same 2-tasks-per-MCP-call rate plus the full task-cost table (Standard/Lead Router/Code/AI tiers): https://zapier.com/pricing/rates (read 22 July 2026)
- Zapier pricing, free plan 100 tasks/mo, Professional 750 tasks/mo at $19.99/mo billed annually, Team 2,000 tasks/mo at $69/mo billed annually, pay-as-you-go overage, Agents and Chatbots tabs: https://zapier.com/pricing (read 22 July 2026)
- AI by Zapier model-tier pricing, Standard 1x / Advanced 3x default / Premium 5x / bring-your-own-key 1x, and the 75-task per-run pause, confirmed current and unchanged since the 15 June 2026 update: https://help.zapier.com/hc/en-us/articles/46425475442829-AI-by-Zapier-model-tier-pricing (read 22 July 2026)
Resolved since 16 July 2026: Zapier's Team tier task allowance, previously UNVERIFIED here, is now published: 2,000 tasks a month at $69/mo billed annually. See the Team row above.
Our arithmetic, not Zapier's: the tool-calls-per-month column, the sessions-per-month column, the 38-task session total, the 798-task working month, the 5.3 cents per call, and the 50-record extension of their "search and update 10 records" example. Every input to those is Zapier's own published figure. Check the division.