Workato

Workato: agents on top of an integration backbone the enterprise may already own

Plenty of agent pilots stall at the point where a clever prototype has to write to the ERP, open a ticket in the service desk and update the CRM in a single, auditable transaction. Workato approaches agentic AI from that end of the problem. Its homepage title promises to govern and scale enterprise AI on what it calls the number one iPaaS¹, and the line beneath it describes a platform that both controls and executes, with three duties named: keeping AI under governance, linking data and apps, and coordinating agents alongside workflows². The documentation repeats the pitch in engineering terms, offering a single orchestration layer where a company can both build its AI agents and MCP servers and keep them under governance³.

This site ranks Workato seventh, and the placement says more about this buyer than about the product. For a company whose stalled pilot needs reliable, governed execution across hundreds of systems, Workato publishes a long certification list and specific data protection controls. It ranks last here because building agents on it is the customer's work, the help on offer is described only as onboarding and training packages available on request⁴, and prices are not publicly stated on the pages we opened.

Genies, recipes and the two planes

Workato's agents have their own name. Agent Studio is where the company says customers build, deploy and govern Genies, its term for AI agents, across the enterprise⁵. Genies carry out actions with several steps across more than 12,000 applications⁶, and because they run on Workato's event driven iPaaS they need nobody to type a request: an API call, a timetable or a business event sets them going⁷. The Agent Studio page says Genies are assembled from more than 750K reusable Recipes and Skills⁸, a vendor figure that points to the platform's heritage: a recipe is Workato's unit of automation, and an agent is built largely by recombining them.

The architecture is described as two layers. The Control Plane governs every interaction between models, agents, workflows, APIs and MCP servers through central identity, policy, approvals, security, observability and cost controls⁹. The Execution Plane provides deterministic orchestration across agents, MCP servers, data, applications and APIs¹⁰. The homepage condenses this into a single line in which the first plane governs each agent and the second does the work¹¹. For a pilot whose failures came from an agent improvising where a fixed process was needed, deterministic orchestration is the relevant design choice.

Two further pieces matter for production. Enterprise MCP turns existing Recipe Functions and Skills into MCP servers without code¹², so automations already in service can be offered to agents as tools. Workbot, the conversational front end, uses natural language processing so users can talk to it in ordinary language¹³. Engineers who need a connector Workato lacks will find the Connector SDK named on the documentation home page¹⁴.

Reaching the systems a production agent must touch

Connection breadth is where Workato's integration history shows most. One number appears on the homepage, which says the platform reaches upward of 17,000 targets spanning MCP servers, files, databases, APIs and apps¹⁵. The figures vary by page, though, and a sponsor should know that before quoting one in a business case: the pricing page counts 1,200 plus prebuilt connectors included for every user¹⁶, the Enterprise MCP page gives agents access to more than 14,000 apps¹⁷, the docs home page also says 14,000 plus¹⁸, and the Agent Studio page says 12,000 plus⁶. The integrations page describes connectors for thousands of SaaS applications, databases and ERPs, both in the cloud and on premises¹⁹.

Several details answer the questions a stalled pilot usually raises. Systems that sit behind the firewall, whether applications, databases or the customer's own software reached through REST or SOAP, are supported through an on premises gateway²⁰, while the security page describes an On Prem Agent that bridges into systems kept on the customer's private network²¹. Model providers are covered by prebuilt connectors for OpenAI, Google Gemini, Amazon Bedrock and Azure OpenAI²². Enterprise MCP works with Claude Desktop, ChatGPT, Workato's own Genies, Cursor and any agent that supports MCP²³, which means agents built on another platform can call Workato automations. Software vendors embedding integrations in their own products can tailor them with the Connector SDK²⁴ and, by Workato's count, reuse any of over 400,000 recipes that a community of more than 70,000 people has published²⁵.

Controls for agents that act on real systems

Workato's governance language is aimed squarely at the moment an agent starts changing records. Granular custom roles define who may create, build and deploy workflows, agents, MCP servers and other assets²⁶. Governance includes human in the loop approvals and fine grained role based access, down to environment and project permissions²⁷. Even the entry edition, Standard, comes with role based access control, custom OAuth profiles, MFA and SSO over SAML²⁸, so the identity basics are present from the lowest tier upward.

Identity is the most distinctive control. On the Agent Studio page, Genies act for verified users and never through shared service accounts²⁹. Enterprise MCP passes the authenticated user's identity through to agent actions, so role based access is enforced and audit trails are kept automatically³⁰, and its access policies route each request according to that user's context³¹. The security page adds that verified user access keeps every person inside the set of applications they hold permission for³². For a security reviewer who blocked a pilot because its agent ran under a powerful technical account, this is the feature to examine first.

Security evidence

Ongoing assurance is described as well. Assurance does not stop at certificates, since the page describes continuous validation through outside penetration testing on a regular basis and a bug bounty that is currently running³³. Every workflow execution, every change to configuration and every user action is written to an audit log whose retention can be set at 90 days³⁴, and the platform masks PII and other sensitive fields³⁵. On the pricing page, Enterprise MCP is the edition that bundles an enhanced security set, listed as runtime user connections, keys rotated hourly, an audit of activity and masking of data³⁶. The Agent Studio page also claims 99.9% uptime³⁷, a vendor figure.

Where data is processed

Data location is unusually well covered. A sponsor whose pilot stalled over residency has eight regions to choose from, Korea, Australia, Israel, China, Singapore, Japan, the EU and the United States, each keeping its data local³⁸. AWS PrivateLink and Azure Private Link keep traffic off the public internet³⁹, and Virtual Private Workato appears as an optional add on⁴⁰, with the security page linking to the deployment architecture and hosting options it belongs to⁴¹. Customers on the Workato Platform plan get to write their own data retention policy, according to an answer in the pricing FAQ⁴², which suits a compliance office with fixed deletion schedules. How long logs survive also depends on the edition bought, a month on Standard⁴³ against a flexible three months on Enterprise MCP⁴⁴, which an audit team should match to its own evidence rules.

A self hosted installation of the whole platform on the customer's own servers is not publicly stated on the pages we opened; the published routes into private networks are the On Prem Agent and the gateway^21,20.

Getting started and getting help

The homepage offers a sign up to start for free⁴⁵. A change board will be glad to see that even Standard ships with three separate environments, one each for dev, test and production⁴⁶. A Usage Dashboard gives visibility and control over consumption⁴⁷, and the Workbot page links a short demo of setting up an enterprise Workbot app in Slack with a custom OAuth profile⁴⁸.

Help with the build is where Workato says least. The only reference we found is a line on the pricing page saying Workato provides a variety of onboarding and training packages and inviting buyers to get in touch for details⁴. A Workato team that designs and builds the customer's first production agent is not publicly stated on the pages we opened. A sponsor whose pilot stalled for lack of builders will need either internal integration engineers or a Workato partner.

Editions, prices and capacity limits

Under the heading Explore our platform editions, the pricing page sets out a ladder of four, with Workato ONE at the top, Enterprise MCP and Business MCP in the middle, and Standard as the entry point⁴⁹. It describes pricing as built for flexibility⁵⁰ and the usage component as flexible, predictable and scalable⁵¹, with consumption tracked by capability, such as Workflows, APIM and Event Streams⁵². Unlimited workspaces and users are listed⁵³, and optional add ons include Advanced Security and Governance and Private Link⁵⁴. Workato notes that plan structures may vary with contract terms and the date a customer joined⁵⁵. List prices for the editions are not publicly stated on the pages we opened.

Capacity figures matter for an agent that will handle production volume. The entry edition is sized for experiments: its allowance is one Standard Workbot⁵⁶, and its concurrency allowance is set at one⁵⁷. On premise connectivity is rationed too, with three OPAs in Business MCP⁵⁸ and ten in Enterprise MCP⁵⁹, while concurrency reaches 30 in Workato ONE⁶⁰. A footnote warns buyers against adding the tiers together, because each figure is the full amount for that edition, never an increment on the one beneath⁶¹. A pilot that ran comfortably on one concurrency may need a higher edition before it can serve a department.

Evidence of use in large companies

Workato claims Fortune 500 companies among its customers⁶², and its homepage puts the number of employees whose agents run through Workato's control and execution plane above 100,000⁶³. The integrations page lists automation examples by function: ticket escalation and threat response in IT⁶⁴, payroll, benefits and onboarding in HR⁶⁵, quote to cash and procure to pay in finance⁶⁶, and lead routing and management in sales⁶⁷. The pricing page carries one customer's estimate of the payoff, roughly 6,500 hours of human effort saved every month⁶⁸. A separate embedded offering targets software vendors that want prebuilt integrations and automations inside their own product⁶⁹.

Verdict for a team with a stalled pilot

Workato ranks seventh for the buyer this site addresses. For a different buyer, an enterprise that already runs Workato as its integration layer and whose pilot stalled on execution and security, it may be a short path: deterministic orchestration¹⁰, agents that act as verified users²⁹, a certification list matched in length here only by Google's⁷⁰, keys rotated each hour⁷¹ and in region hosting across eight geographies³⁸.

It ranks last here because the customer, or a partner, builds the agents, delivery help is limited on the pages we opened to onboarding and training packages on request⁴, edition prices are not publicly stated, connector counts differ across pages^15,16,17, and the lower editions carry tight concurrency limits⁵⁷. A sponsor should treat Workato as a strong execution and governance layer that still needs a team to design and build the agent on top of it.