Zapier connects roughly nine thousand business applications so that work moves between them automatically, and now positions itself as the control plane between AI models and the tools they need to act on. It suits operations, marketing and IT teams who want automation without engineering — provided they understand that billing runs on successfully completed tasks.
Used this tool? Rate it
Used this tool? Rate it
Zapier is an automation platform whose core function is to connect business software that does not otherwise talk to each other. When something happens in one application — a form is submitted, a payment clears, a deal changes stage — Zapier notices and makes something happen in another application. It has been doing this for over a decade and is the most widely adopted product in its category.
What it is today is broader than that description suggests, and the gap matters because most directory entries and even many recent reviews still describe the 2018 version of the product. The current homepage positions it as something considerably more ambitious: "Build and govern AI workflows and agents across 9,000+ apps with Zapier — the control plane that connects any AI to any tool". The company reports serving over 3 million businesses, a vendor-stated figure rather than an audited one.
That repositioning is not merely marketing. The product line named on the homepage now comprises Zaps, Agents, Canvas, MCP and the Zapier SDK. Only the first of those existed in the classic conception of the tool. The strategic argument is straightforward: Zapier spent a decade building authenticated connections to thousands of applications, and in an era where AI models need to take actions in real systems, that connection layer is arguably more valuable than it ever was as a standalone automation product.
Corporate identity is fully documented, as one would expect at this scale. The terms of service state that "Zapier, Inc. is a Delaware corporation, and in these Terms, we will reference ourselves as "Zapier" or "we"/"us"." and that the agreement "will be governed by and construed in accordance with the laws of the State of Delaware, without reference to its conflict of laws principles."
If you know Zapier only as "the thing that connects app A to app B", update that model before evaluating it. The tags under which it is often catalogued — API, interfaces — capture a fraction of what it now does. Understanding the current product means understanding four distinct billing and capability surfaces, not one.
A Zap is a trigger paired with one or more actions. Something happens; something else follows. Multi-step Zaps chain several actions together, add filters and branching, and transform data between steps. This remains the backbone of the platform and the reason most users arrive.
Note one structural constraint at the entry level: the free tier is limited to "two-step Zap workflows", meaning one trigger and one action. Multi-step automation — which is where most real business logic lives — requires a paid plan.
Agents represent the platform's answer to the shift from deterministic automation to reasoning systems. A Zap follows a path you defined in advance; an agent decides what to do within boundaries you set. The critical practical fact, covered in detail under pricing, is that agents are metered differently from Zaps, in "activities, not tasks".
This is the most strategically significant recent addition. Zapier MCP exposes the platform's application connections to outside AI clients through the Model Context Protocol. The product page frames it succinctly: "Your AI can talk. Zapier MCP makes it act." It names Claude, ChatGPT and Cursor as supported clients and covers the same 9,000+ application catalogue.
The security model deserves attention because it addresses the obvious objection to letting a language model touch production systems. Per the product page, "MCP uses the same credential infrastructure Zapier has managed for 13+ years", and crucially: "You control which apps your AI can access and which actions are permitted. Every action is logged in your History tab. Zapier maintains SOC 2 Type II compliance." Per-app and per-action permission granularity plus a complete audit log is a materially better governance story than handing an AI assistant broad API keys.
The platform includes its own lightweight database and form builder, and notably both are unlimited even on the free plan: "unlimited Zap workflows/Tables/Forms". This matters more than it sounds. Many automations fail not because the logic is hard but because there is nowhere to store intermediate state. Having a native data store removes a common reason to bolt on another subscription.
Canvas provides a visual space for mapping processes before building them. Copilot assists with construction and is available on the free plan subject to a daily limit. Both target the same problem — that the hardest part of automation is usually designing the process, not wiring the steps.
Beyond the no-code product there is a Zapier SDK, and the company runs a developer platform program which, per its privacy statement, "allows third-party application providers to build integrations to Zapier". This is the mechanism behind the app catalogue: much of it is built by the application vendors themselves, which is how the number reached nine thousand.
The classic and still the most common. A form submission or ad lead flows into the CRM, triggers a notification, adds a task and updates a spreadsheet — without anyone copying data between systems.
Routing events from one system into the place a team actually watches, whether that is a chat channel, an email inbox or a ticketing queue. Simple to build and immediately visible in value.
Keeping a CRM, a billing system and a support desk consistent with each other. This is unglamorous, high-value work that otherwise consumes hours of manual reconciliation.
Order events driving fulfilment updates, customer messages, accounting entries and inventory adjustments.
Publishing pipelines, social distribution, campaign reporting and the many small handoffs between marketing tools that otherwise require someone to remember them.
The newest and fastest-growing use case. Through MCP, an AI client can send an email, update a CRM record or create a ticket rather than merely describing how to do so. For teams already building with Claude, ChatGPT or Cursor, this converts a conversational assistant into something that operates on real systems, with permissions and an audit trail.
Tables plus Forms plus Zaps together cover a surprising amount of what small internal applications actually do: capture input, store it, act on it, notify someone.
The best first automation is a task you personally perform several times a week and resent. Automating something rare produces little value and teaches you little.
This is the single most important thing to learn about Zapier, because it governs your bill. The definition is precise: "A task is counted whenever Zapier successfully completes a unit of work for you. Failed actions are not counted." Every action step that succeeds consumes one task, per item processed.
Equally important and frequently misunderstood: "What does not count: triggers, polling for new data, and Zapier built-in data tools." Zapier checking for new data costs nothing. Filters and formatters cost nothing. Only successful actions cost. This single fact should shape how you design workflows.
Because triggers and filters are free while actions are metered, the economical design puts filtering as early as possible. A Zap that triggers 1,000 times, filters down to 20 relevant items and then acts costs 20 tasks, not 1,000. Getting this wrong is the most common cause of unexpectedly large bills.
Prove that the trigger fires and the action lands before adding logic. On the free plan you are limited to two steps anyway, which is enough to validate the connection.
When an automation needs to remember something between runs, use the built-in Tables rather than reaching for another service. Built-in data tools do not consume tasks.
Agents cost differently and are worth reaching for when the decision cannot be expressed as a rule. Where a filter suffices, a filter is cheaper and more predictable.
If you are exposing tools to an AI assistant, grant the narrowest scope that works — read-only where reading suffices — and review the History log after the first days of use to see what the model actually did.
Once you internalise that triggers and filters are free and successful actions are metered, a lot of design decisions become obvious. Filter aggressively, batch where possible, and avoid actions that fire on items you will discard.
Task consumption scales with data volume, not with the number of Zaps. One automation processing a high-traffic source can consume more than dozens of low-volume ones. Review consumption after the first full month rather than assuming your estimate held.
This is a contractual requirement, not just prudence. The terms prohibit submitting payment card and financial account numbers, government identification numbers, and GDPR special-category personal data. Design workflows to pass references rather than the sensitive values themselves.
The terms permit Zapier to create de-identified Derived Data from customer content and use it for product development "including through model training", with an opt-out form available. Whether to exercise that is a policy decision your organisation should make consciously rather than by default.
The ability to grant read-only or specific-action access is the main safety mechanism when connecting a language model to production systems. Use it rather than granting broad access for convenience.
An operational trap worth knowing: per the terms, an email, chat or phone request does not constitute cancellation, nor does removing your payment method. Cancellation must be performed in the account itself.
Third-party reviews consistently note that beginners find even simple automations initially overwhelming. Budget an hour of orientation rather than concluding after ten minutes that it does not fit.
Deterministic rules are cheaper, faster and debuggable. Agents earn their cost where the input genuinely varies in ways rules cannot capture.
The core audience. People responsible for the plumbing between systems, who need results without a development cycle.
Campaign, lead and content workflows span many tools, and marketing rarely has dedicated engineering support.
The proposition is strongest here: integrations that would otherwise require custom code become configuration.
For the long tail of small integrations that never justify a project, plus governance features and SOC 2 Type II compliance for organisations that need them.
MCP makes this platform one of the most direct ways to give Claude, ChatGPT or Cursor the ability to act in real business systems with scoped permissions and logging.
Teams with very high-volume, low-value events, where per-task pricing works against you and a code-based integration is cheaper. Organisations that must process sensitive data categories the terms prohibit. Engineering teams who would rather own the integration code. And anyone needing complex conditional logic beyond what the builder expresses comfortably — a limitation reviewers raise regularly.
Zapier is a browser-based service; automations run on its infrastructure rather than on your machine, which means workflows continue when your computer is off. That is the fundamental architectural point, and it is why it can serve as connective tissue between systems that never interact directly.
The platform surface extends in three directions beyond the web editor. The Zapier SDK supports programmatic construction. MCP exposes the connection layer to external AI clients, making the platform usable from inside Claude, ChatGPT or Cursor rather than only from Zapier's own interface. And the developer platform program lets third-party software vendors build their own integrations into the catalogue — the mechanism that produced a library of roughly nine thousand applications, since the vendors themselves do much of the work.
Around this sit ecosystem programs described in the privacy statement, including "our experts program (which connects experts who help our customers set up and use the Services), our partner reseller program (where authorized partners may resell our Services), our partner referral program". For buyers this matters practically: implementation help is available from a certified third-party market rather than only from the vendor.
Zapier's pricing is straightforward in structure and easy to misjudge in practice, so it is worth being precise.
The billing unit is the task, defined as follows: "A task is counted whenever Zapier successfully completes a unit of work for you. Failed actions are not counted." The complement matters just as much: "What does not count: triggers, polling for new data, and Zapier built-in data tools." Zapier watching for events is free; only completed actions are billed. That asymmetry is the key to controlling cost.
The free plan includes 100 tasks per month along with unlimited Zaps, Tables and Forms, two-step workflows, Copilot with a daily limit, and basic access to AI products. It is genuinely usable for personal automation and for proving a concept before spending.
Paid plans scale by task volume rather than by feature gates alone. Professional begins at 750 tasks per month for $19.99 monthly on annual billing ($29.99 month-to-month) and climbs through many tiers to two million tasks per month at $3,389.00 on annual billing. Team starts at 2,000 tasks for $69.00 on annual billing and scales along a similar curve, adding collaboration capabilities. Enterprise is custom-quoted through sales.
Agents introduce a second meter, and this is the part most likely to surprise existing users: the "Billing unit: activities, not tasks", where an activity is "any billable action an agent takes", with 400 per month included at no cost. If you plan to use agents at scale, model that consumption separately from your task budget.
Two commercial terms to note. Fees are "non-refundable, and payment obligations are non-cancelable" except as stated or legally required, and anything provided during a free trial is supplied as-is without warranties. Given the volume-based pricing, this argues for starting on a lower tier and measuring real consumption for a month before committing.
Make (formerly Integromat). The most direct comparison. Its visual builder handles complex branching more comfortably, and its operations-based pricing often costs less at high volume. Zapier generally leads on integration breadth and on the polish of individual connectors.
n8n. Source-available and self-hostable, which changes the economics entirely for high-volume work and satisfies teams that must keep data in their own infrastructure. It demands more technical capability to run.
Platform-native automation. Microsoft Power Automate within the Microsoft ecosystem, or the workflow features already inside your CRM or project tool. If your integrations stay within one vendor's ecosystem, the native option is usually cheaper and better integrated.
Custom code. For a small number of very high-volume integrations, direct API code eliminates per-task costs. The trade is that you now own maintenance forever, which is exactly the cost Zapier exists to remove.
The honest framing: Zapier competes on catalogue breadth, reliability and time-to-first-working-automation. It is rarely the cheapest option at very high volume, and it does not try to be.
The most consistent criticism in independent reviews. Because pricing follows task consumption, growth in business activity translates directly into a larger bill, and teams frequently move up several tiers faster than expected. Model your realistic volume before committing, and design filters to keep task counts down.
Reviewers on G2 repeatedly raise restrictions on workflow complexity alongside plan-based usage caps. Deeply branching, conditional processes are possible but become awkward to build and maintain compared with tools designed around visual branching.
Despite the no-code positioning, third-party reviews note that newcomers find even simple automations overwhelming at first. The concepts are learnable, but expect a genuine learning curve.
A hard constraint that catches teams mid-project: the terms bar submitting "credit, debit, bank account or other financial account numbers", "social security numbers, driver's license numbers, or other government ID numbers", and "special categories of personal data enumerated in the European Union General Data Protection Regulations". Any workflow touching these needs redesign or a different tool.
Start from the definition, because it is broader than most people assume: the terms define Customer Content as "any information, content, or materials that you submit through the Service to your Zapier Account, or is generated and returned by the Service based on such submissions, including Inputs and Outputs." That covers both what you send and what comes back.
On top of that, the terms permit Zapier to build de-identified Derived Data from customer content and use it "to operate, enhance, improve, and develop Zapier products or services, including through model training", with an opt-out form provided. The opt-out is real and welcome; the default is participation.
Standard for SaaS but worth stating: "The Service is made available on a limited access basis, and no ownership rights are conveyed to you", with Zapier and its licensors retaining all intellectual property in the platform. Your automations run on someone else's system under their terms.
Requests by email, chat or phone do not cancel a plan, and neither does revoking your payment method. Cancellation must happen in the account. Zapier also reserves the right to suspend or terminate accounts.
Combined with volume-based pricing, this makes conservative tier selection the sensible approach.
The privacy statement is explicit that the sites and services are not directed at or intended for children under 16, and that information is not knowingly collected from them.
The application count of roughly nine thousand and the figure of over three million businesses come from Zapier itself. They indicate scale but are not independently audited.
A task is counted whenever Zapier successfully completes a unit of work, and failed actions are not counted. Crucially, triggers, polling for new data and Zapier's built-in data tools do not count. So Zapier watching for events and filtering them costs nothing; only the actions that complete are billed. Design workflows to filter before acting and your costs drop accordingly.
Yes, within limits. It includes 100 tasks per month, unlimited Zaps, Tables and Forms, Copilot with a daily cap and basic AI access. The real constraint is that workflows are limited to two steps — one trigger and one action — so multi-step business logic requires a paid plan.
The current official figure is more than 9,000. Be aware that many third-party articles still quote 6,000 or 8,000, which reflects older snapshots rather than a discrepancy — the catalogue grows continuously because third-party vendors build their own integrations through the developer platform.
Agents use a separate meter: the billing unit is activities rather than tasks, where an activity is any billable action an agent takes, with 400 included per month at no charge. If you intend to use agents heavily, budget that consumption separately from your task allowance.
MCP exposes Zapier's application connections to external AI clients — Claude, ChatGPT and Cursor are named — so that an assistant can actually perform actions across the app catalogue instead of only describing them. It uses the same credential infrastructure Zapier has run for over a decade, lets you scope access per app and per action, logs every action in your History tab, and operates under SOC 2 Type II compliance.
Partly, unless you opt out. The terms allow Zapier to create de-identified Derived Data from your content and use it to develop its products, including through model training, and an opt-out form is provided. Treat that as a decision to make deliberately.
Yes, and this is contractual rather than advisory. Financial account and payment card numbers, government identification numbers such as social security or driver's licence numbers, and GDPR special-category personal data are all prohibited. Redesign affected workflows to pass identifiers or references instead.
Cancel inside your account — an email, chat or phone request does not count as cancellation, and neither does removing your payment method. Fees paid are non-refundable and payment obligations non-cancelable except where the terms or law provide otherwise, so choose your tier conservatively.
Zapier, Inc., a Delaware corporation. The terms are governed by the laws of the State of Delaware, with disputes submitted to the exclusive jurisdiction of courts there. Its privacy statement is issued by Zapier, Inc. and its affiliated entities.
Two themes dominate independent reviews. Cost escalation as volume grows, which follows directly from per-task pricing, and limits on workflow complexity together with plan-based usage caps. A secondary theme is that the interface is less approachable for beginners than the no-code framing implies. None of these are hidden defects — they are the trade-offs of the model — but they should inform your evaluation.
