Celigo Ora and Agent Builder: What NetSuite Teams Need to Know

If you run NetSuite on Celigo, the same bottleneck probably sounds familiar. A finance team waits days for a mapping change that takes an hour of actual work, because the one person who understands the flow is booked on something else.

Errors sit in a queue nobody fully owns. A flow built correctly two years ago quietly drifts out of sync with the business and nobody notices until month end goes wrong.

Celigo built three related capabilities aimed at that gap: Ora, Agent Builder, and an MCP server. They are not the same thing, and most coverage of the launch has blurred the difference. This guide assumes you already know what Celigo is and are using or evaluating it for NetSuite.

It separates the three capabilities, sets out what each can actually do today, and covers the governance decisions that determine whether opening them up helps your estate or adds a new kind of mess.

What Celigo Ora Is

Ora is a natural language interface built into Celigo itself, not a chatbot layered on top of one screen. You describe what you want changed, built or diagnosed, and a set of specialised agents that understand your account’s connections, flows and data act on the request.

Celigo describes Ora as a multi-agent system: separate specialists for flows, connections, integration error management, scripts and APIs, coordinated by an orchestration layer that routes each request to the right one. Connect it to your account and it builds a live knowledge graph of every integration, flow and dependency, which is what lets it reason about the impact of a change before making it.

Nothing happens automatically. Ora stages every change for review, and offers two modes: have it act immediately, or have it walk you through the plan first. That approval step is the same design principle that runs through governance across all three capabilities, covered in more detail below.

Ora, Agent Builder and MCP: How They Differ

These three ship together but solve different problems. Confusing them leads to the wrong adoption plan.

Put simply: Ora is how you work on your Celigo account. Agent Builder is how you create something that runs as an automation on your account. The MCP server is how you let AI agents outside Celigo call into your account’s capabilities without giving them the keys.

What Each Capability Can Do Today

Ora

Celigo documents four practical things Ora does: building new flows from a plain-language description, modifying existing automations through the same interface, diagnosing why something failed by tracing audit logs and dependency data to a root cause, and operating automations day to day without menu navigation.

It works across deterministic flows, AI agents, B2B flows and data syncs through one interface, and covers roughly twenty resource types, including flows, connections, scripts, APIs and EDI profiles.

One design detail matters for cost and reliability: Celigo’s platform decides per step whether an automation should run on a deterministic script or use generative AI for genuinely ambiguous work such as classifying intent. A validated order moving from Shopify into NetSuite is not ambiguous, and it is worth confirming during setup that your financial flows are running deterministically rather than through a model.

Agent Builder

Agent Builder lets you define an agent’s instructions, the tools and connections it can use, and its multi-step logic, then attach guardrails so it stays inside policies you set. You can require human approval at any decision point you consider critical.

Celigo’s guardrails can screen for PII exposure, apply content moderation and enforce custom business rules on what an agent is allowed to do. Support for Google Gemini models alongside OpenAI, including bring-your-own-key configurations, was added in a later 2026 release.

Celigo MCP Server

You build your own MCP server inside Celigo’s AI Studio, choosing which APIs and integration capabilities to publish as callable tools, with an authentication token controlling access and full request logging for audit.

A later release added support for MCP resources (read-only reference data agents can pull for context) and MCP prompts (reusable, standardised prompt templates), plus the option to authenticate connecting clients with a user’s own Celigo sign-in instead of a separate identity provider.

Celigo names specific tools in its own documentation for that identity-based sign-in flow: users connecting through Claude or Cursor are directed to a standard Celigo sign-in page.

Celigo’s broader compatibility claim is that any MCP-compatible assistant, IDE or agent framework can invoke the server, citing OpenAI and Claude as examples of where an agent might be built. We are not extending that list to other clients Celigo has not named, since behaviour against an unlisted client is unverified.

Practical Use Cases for a NetSuite Team

Be specific here, since the vendor framing is broad and most teams have a narrower budget than the marketing suggests.

Error triage and first-line diagnosis. Integration errors follow patterns: a required NetSuite field is empty, a subsidiary was not assigned, a customer record does not exist yet, or an API rate limit was hit at a peak hour. Describing a failure and getting a traced diagnosis back saves a senior person’s attention on a routine problem, which is usually the most expensive resource in an integration operation.

First drafts of standard flows. Order sync, customer sync, item sync and fulfilment updates against well-documented APIs are where natural language building is strongest. The output still needs review. Be more cautious with anything touching revenue recognition, multi-book accounting, intercompany transactions or custom SuiteScript hooks, where the accounting consequence of a mistake is the real risk, not the field mapping.

Reducing the change queue for business teams. If operations or finance currently files a ticket for every threshold adjustment or notification tweak, a governed self-service path removes real friction, provided the governance is configured first.

A narrow agent for one bounded task. Agent Builder is a better fit for a well-defined, repeatable job with a clear success measure, such as classifying inbound support tickets or routing exception records, than for open-ended automation of a whole process.

What Remains Under Human Control

Governance is built into the platform rather than bolted on afterward. Every Ora action is staged for approval before it takes effect. Agent Builder lets you require a human checkpoint at any step you name. Celigo maintains full audit logging across MCP server requests, Ora actions and agent executions, recording inputs, decisions and resulting changes.

What the platform does not do is decide these boundaries for you. Which flows can run unattended, which need a person to approve first, and who is allowed to build against which connections are decisions your team has to make deliberately, account by account and flow by flow.

Permissions, Data Access and Governance

Before opening any of this beyond your integration specialists, settle the following.

Which flows require human approval. Anything writing to the general ledger, touching revenue recognition, changing item costing or affecting a customer-facing commitment should sit behind an approval gate, configured per flow rather than once at account level.

What an agent is permitted to do autonomously. For Agent Builder specifically, the operative question is not what an agent can technically do but what it is allowed to do without asking first. Start narrower than feels necessary and widen based on evidence from real runs.

Environment discipline. Nothing about natural language building changes the case for building and testing in sandbox first. Faster building makes environment separation more important, not less.

MCP token scope. Each MCP server’s access token defines what an external agent can reach. Treat token issuance with the same care as any other credential with write access to business systems, and review the request log periodically rather than assuming the audit trail is being watched.

Which Option Fits Which Organisation

Not every team needs all three capabilities on day one.

Readiness Checklist

• Confirm your Celigo edition includes Ora, Agent Builder and MCP server access, and check current entitlement with your account team.
• Identify who on your team will have build access initially, and to which connections.
• List which flows touch the general ledger, revenue recognition, item costing or customer-facing commitments, so approval gates can be scoped correctly.
• Confirm your NetSuite master data (subsidiaries, item records, custom fields) is clean enough to build against safely.
• Decide your sandbox-first policy for anything built through Ora or run through Agent Builder.
• If considering the MCP server, identify which external AI tools would actually consume it and what data they need access to.

A Practical Adoption Path

• 1

  Weeks 1 to 2. Turn on Ora for your existing integration team only, and use it exclusively for diagnosing errors in your live open error queue. Low risk, measurable, and it teaches your team how the tool reasons before you trust it with builds.

• 2

  Weeks 3 to 4. Add build access in sandbox. Have your team rebuild a flow they already know well and compare the result to what they would have built manually, to calibrate trust rather than save time yet.

• 3

  Weeks 5 to 8. Define the governance model in writing: which flows need approval, who can build against which connections, what the review cadence is. Then extend access to one business team with a genuine backlog of small requests.

• 4

  Month 3 onward. Evaluate Agent Builder for one narrow, well-bounded use case with a clear success measure. Avoid anything with direct financial posting consequences until you have operational evidence from lower-stakes work.

The same readiness questions that apply to enabling AI features inside NetSuite itself apply here. Our NetSuite AI readiness checklist covers that groundwork in more depth.

Where This Leaves a Celigo and NetSuite Team

Ora, Agent Builder and the MCP server change where time gets spent on integration work, not whether someone needs to understand the estate. Diagnosis and first drafts get faster. Architecture decisions, NetSuite configuration judgment and ownership of how flows interact with each other still benefit from experience, and still need a name attached to them.

The teams that get real value from this early are the ones that scope access deliberately and treat governance as a design decision made before rollout, not a setting revisited after something goes wrong.

Primary Sources

Facts in this article are drawn from Celigo’s own product pages and Help Center documentation, and from independent reporting on the general availability announcement, cited beside the claims they support above. Celigo-reported adoption figures are identified as vendor-reported where used.

• Celigo, “Celigo Ora” product page
• Celigo, “Build AI Agents for Enterprise Workflows”
• Celigo Help Center, “MCP server overview”
• Celigo Help Center, “Learn about Celigo AI agents”
• Celigo Builders Hub, “Celigo April 2026 Release”, published 13 April 2026
• Celigo Help Center, “Celigo agentic automation January-May 2026 release notes”
• SiliconANGLE, “Celigo launches intelligent enterprise automation builder into general availability”, 2 September 2026
• Celigo, “Celigo Agent Builder + Ora — Build AI Agents Across Every System” (product video), YouTube, @CeligoInc