What can AI do inside NetSuite, and how does the AI connector work?
Short answer: NetSuite, Oracle's cloud ERP (the main system for orders, stock, purchasing and accounts), has two kinds of AI. It has built-in features such as invoice capture, anomaly detection and text drafting. It also has the AI Connector Service, which lets an AI assistant you already use, such as Claude or ChatGPT, read and update NetSuite data through a standard called MCP. Oracle's FAQ says the connector is not a paid feature, though your AI plan may cost money, and it runs under a non-administrator NetSuite role.
You run NetSuite, you keep hearing that AI can now answer questions from your own data, and you want to know what is real, what it takes to turn on, and where it can hurt you.
I'm Ranjit, and I run Blur Studio: 15+ years building software, 120 apps taken live, and clients who have gone on to raise $10M+. I build AI systems and processes that help businesses do a lot more with a lot less time. I also work differently from most of what else is out there, more on that further down this page. Say hello.
If you want to know where AI would pay off in your own distribution business before you connect anything, the AI Strategy Sprint finds it in 3 days.
What AI does NetSuite have built in?
NetSuite's AI page splits its features into assistants that do a task and advisors that explain data. The ones it names are:
- Invoice capture: reads scanned supplier invoices so staff type less.
- Anomaly detection: NetSuite Exception Management scans financial data and flags unusual entries.
- Text Enhance: uses generative AI (AI that writes text) to draft and refine content inside NetSuite, with Prompt Studio to set tone and format.
- Narrative reporting: writes plain explanations from financial and transaction data.
- Enriched bank data: helps match imported bank lines to ledger entries.
- NetSuite Expert: an AI agent that answers questions from NetSuite support material.
- Intelligent item recommendations: suggests products during a sale based on what the customer bought before.
- Data discovery: NetSuite Analytics Warehouse answers questions about your data in conversation.
The page gives no price for these. It links to a resource on AI units, which is how NetSuite measures AI use. Ask your account manager which features your contract includes and whether any use is metered.
What is the NetSuite AI Connector Service?
This is a different thing. NetSuite calls it a bring-your-own-AI route. Instead of using NetSuite's AI, you let your own AI assistant talk to NetSuite. It uses MCP, short for Model Context Protocol, an open standard that lets an AI assistant call tools in another system. If you want the plain-language version of what an AI agent is, I wrote a guide to what an AI agent is.
By default, Oracle says the MCP Standard Tools SuiteApp (a free add-on package you install from NetSuite) comes with tools for working with records, reports, saved searches and SuiteQL queries. SuiteQL is NetSuite's SQL-style query language, and Oracle says the SuiteQL tools support read-only queries.
When you ask a question, the AI picks a tool, the tool asks NetSuite, NetSuite answers under your permissions, and the AI writes the result in plain words.
Which AI tools work with it, and what does it cost?
Oracle's FAQ says any client that supports remote MCP, protocol version 2025-06-18, streamable HTTP and OAuth 2.0 with PKCE (a sign-in method) can work. It names Claude Pro and ChatGPT as currently supported, and says some ChatGPT plans need Developer Mode switched on.
On cost, the FAQ says the connector is not a paid feature and the MCP Standard Tools is a free SuiteApp. You may still need a paid subscription for your AI client. That is the real cost for most owners: a paid AI plan, plus an administrator's time.
How do I set it up?
Have your NetSuite administrator do this. Oracle's FAQ lists the steps and requirements:
- Enable Server SuiteScript, OAuth 2.0 and REST Web Services in NetSuite.
- Install the MCP Standard Tools SuiteApp.
- Create a custom role, or use an existing non-administrator role, with the MCP Server Connection and OAuth 2.0 Access Tokens permissions. Some tools also need REST Web Services.
- In your AI client, add a connector that points to your NetSuite address, which ends in /services/mcp/v1/all. Oracle says that without the /all ending the connection looks disconnected.
- Sign in through NetSuite and accept the data-sharing notice.
Oracle says the connector does not support the Administrator role. That is a good rule, not a nuisance.
What to check if it does not connect
Oracle's FAQ also lists the usual causes. The connection needs an integration record (the NetSuite entry that lets an outside app sign in) with Public Client, a redirect address, the OAuth authorization code grant and the AI Connector Service scope all switched on. It may be created automatically and sit as pending until you enable it. The permission is called OAuth 2.0 Access Tokens, not Access Tokens, and mixing them up is a common slip. If the AI client shows connected but no tools appear, check that the SuiteApp is installed, that you signed in with the right role, and that the address ends in /all. Oracle says that to fetch tools from one SuiteApp only, you use a different address ending in /suiteapp followed by that app's ID. It also says ChatGPT may need a new integration record for each connection, because its callback address changes. When it works, tools appear in the client and the execution log shows activity.
What tools does the standard SuiteApp include?
Oracle's tool list names 25 tools in seven groups. Knowing them helps you set a role that allows only what you want.
- Records (6 tools): create, get, update, and transform a record into another supported type, plus two that describe record types and supported transformations.
- Record actions (2): find the actions available on a record, then run one.
- Reports (6): list reports, run a report, and list subsidiaries, accounting books, accounting contexts and nexuses to use as filters.
- Saved searches (2): list and run them.
- SuiteQL (2): read the table layout, then run a read-only query.
- File cabinet (6): attach or detach a file on a record, find files by name or path, and list a folder. Detaching does not delete the file.
- Session (1): shows the account, role, features and permissions of your current connection.
The session tool is handy on day one, because it shows what the AI can actually do under the role you made. Oracle says a tool appears only if your role has its permissions, and the first thing to note is that the create, update, transform and action tools are the ones that change data. A read-only role simply leaves them out of reach.
How reports behave through the connector
Oracle's report page says the tools run most standard and custom reports with the filters and footer options from the NetSuite screen, within the role's access. With subsidiaries, a report uses the chosen subsidiary's currency, and a negative subsidiary ID means a consolidated view of that subsidiary and its children. Three limits matter. Accounting periods are ignored, so the AI may assume your fiscal year is the calendar year. Custom filters are not supported. Custom reports that split columns by custom segments may return empty values. So a month-end number the AI gives you can differ from the one on your screen. Check it before you act.
What can the connector do?
Oracle's FAQ says you can create, read and update records, run reports, run saved searches and run read-only SuiteQL queries. You can also build custom tools, which takes a SuiteScript 2.1 script, a JSON schema and a toolset object. That is developer work, and Oracle suggests custom tools when you need specialized workflows, faster single-step operations or custom automation.
What are the limits and risks?
These come straight from Oracle's FAQ.
- Role limits: Tools can only see what the role can see. They cannot run as an administrator, call outside APIs or run elevated scripts.
- Report filters: Reports support only date and subsidiary filters. Accounting periods are not supported.
- Row limits: AI clients usually handle up to about 5,000 rows per call, and large queries can time out.
- Shared request limits: NetSuite's concurrency limit is shared with your other integrations. Too many requests fail with a "Too Many Requests" error.
- Data leaves NetSuite: Data sent to the AI company is governed by that company's privacy policy. Oracle says it cannot control what external components do with data after it leaves.
- Wrong answers: Oracle says AI clients may still produce incorrect or made-up results, so verify against NetSuite data.
- Health data: Oracle says the service has not been assessed for HIPAA (the US health privacy law). Do not send protected health information through it.
- Logs: The execution log keeps 21 days in production and 7 in the sandbox, so review it on a schedule.
Rules Oracle suggests for safe use
Oracle's FAQ recommends using only approved AI clients, limiting permissions, avoiding high-privilege roles, restricting tools for each agent, reviewing access regularly, watching the logs and training users. It also suggests short-lived sign-in tokens with narrow scope, and choosing AI clients that show you the tool name and what it will do before you approve. In a small company, that comes down to one person owning the role, a short list of approved users, and a monthly look at the execution log.
How can a distributor use it?
The good uses are questions that already live in NetSuite but take someone ten clicks to answer. Here are four that fit a distributor or small manufacturer.
- Stock and reorder questions: "Which items have fewer than 3 weeks of stock at the last 90 day sales rate?"
- Customer questions: "Which of my top 50 accounts have open orders past their ship date?"
- Collections: "List invoices over 45 days late, grouped by sales rep."
- Prep for calls: "Summarize this customer's last 12 months of orders before I phone them."
Here is a made-up example. A rep is about to call an account. Instead of opening four screens, the rep asks the AI for that account's orders, returns and open balance. The AI runs a saved search and a report, and answers in a short summary. The rep checks the three numbers that matter against NetSuite and makes the call.
What I would not do yet is let the AI change prices, credit limits or inventory records on its own. Start with a read-only role. Add change rights only for one narrow job, and keep a person approving each change. My guide to AI workflow explains how to keep a human check in the loop.
When is the connector not enough?
It is built around what NetSuite holds. It does not read an emailed purchase order, check your customer portal, or blend NetSuite with your web store. For those jobs you need a link built for your process. My page on AI integration services covers that work, and my guide to NetSuite reports and integrations covers the non-AI options. If you are deciding which first job to try, the AI Strategy Sprint is a paid 3-day first step where you and I find where AI could transform your business, and you leave with the exact steps written down, starting with a free 15-minute AI brainstorm call.
How I'd approach this
I would turn the connector on in a sandbox first, with a read-only role and one person using it. I would pick three questions the team asks every week, run them through the AI, and check each answer against NetSuite for two weeks. If the answers hold up, I would widen access slowly.
My rule: the AI drafts, a person decides. The connector is a good, cheap way to learn what AI can do with your data, because Oracle does not charge for it. It is not a replacement for a clean process. If your pain is orders arriving by email or customers phoning for stock, the connector will not fix that, and I would look at a purpose-built tool instead.
Questions people ask
What is the NetSuite AI Connector Service?
It is a service that lets an AI assistant such as Claude or ChatGPT use NetSuite data through the MCP standard. It runs under a NetSuite role, so the AI sees only what that role allows.
Is the NetSuite AI connector free?
Oracle's FAQ says the connector is not a paid feature and the MCP Standard Tools SuiteApp is free. You may still need a paid plan for your AI client, so check that plan's price.
Can Claude connect to NetSuite?
Oracle's FAQ says Claude Pro and ChatGPT are currently supported. Your NetSuite administrator creates the role and sets up the connection, and the sign-in uses OAuth 2.0.
Can the AI change NetSuite records?
Oracle says the standard tools can create, read and update records, within the role's permissions. I suggest starting with a read-only role and adding change rights for one narrow job at a time.
Is it safe to give AI access to NetSuite?
It can be, if you limit the role, avoid administrator access, review the logs and treat answers as drafts. Oracle notes that data sent to the AI company falls under that company's privacy policy, and that the service is not assessed for HIPAA.
Does NetSuite have AI agents of its own?
NetSuite's AI page lists NetSuite Expert, an AI agent that answers questions from NetSuite support material, plus features such as invoice capture and anomaly detection. Ask your account manager what your contract includes. My AI agent guide explains the term.
More in this topic: B2B: ERP reporting and integrations (10 guides) →
Sources
- NetSuite AI Connector Service FAQ (Oracle NetSuite Help)
- Artificial Intelligence in NetSuite (NetSuite)
- Executing SuiteQL Queries Through REST Web Services (Oracle NetSuite Help)
- Available Tools in the MCP Standard Tools SuiteApp (Oracle NetSuite Help)
- Report Tools in the MCP Standard Tools SuiteApp (Oracle NetSuite Help)
- SuiteAnalytics Connect (Oracle NetSuite Help)