In plain English
The NetSuite AI Connector Service lets an AI assistant like Claude or ChatGPT talk to your NetSuite account safely. Instead of logging in and clicking through menus, you ask a question in plain English — “what are our top 10 overdue invoices?” — and the AI fetches the answer from live NetSuite data. It works through an open standard called MCP (Model Context Protocol), so you can bring the AI tool you already use rather than being locked into one vendor. Crucially, the AI can only see and do what your role allows — the same permissions that protect the regular NetSuite screens protect the AI too.
Same data, a faster front door. The connector turns “dig through NetSuite” into “ask a question” — while keeping your existing permissions in charge.
Why does this matter?
Because the people who need NetSuite answers most — sales leads, finance managers, support agents — often don’t want to learn saved searches and SuiteQL. The connector lets them ask. And because it’s built on an open protocol, you’re not betting your company on one AI vendor: today it’s Claude or ChatGPT, tomorrow it could be something else, using the same plumbing. The flip side is that pointing AI at your ERP needs real guardrails — which is exactly why NetSuite makes you opt in, role by role.
Setting it up, in five steps
You don’t just “turn on AI.” A safe setup is a sequence: switch on the right features, install the tools, build a tightly-scoped role, connect your AI client, then test and watch the logs.
Enable → install → scope a role → connect → test & govern. The expert version and the step-by-step companion have the exact features, permissions, and URLs.
The jargon, in plain English
A handful of terms unlock the whole thing:
MCP
Model Context Protocol — the open “language” that lets an AI client discover and call tools in NetSuite.
AI Connector Service
NetSuite’s built-in MCP server — the secure door an AI client knocks on.
MCP Standard Tools SuiteApp
A free install giving pre-built tools for records, reports, saved searches, and SuiteQL.
AI client
The assistant you connect — Claude, ChatGPT, or another MCP-capable app.
Integration record
The NetSuite record that holds the OAuth connection. Often auto-created on first connect.
Server URL
The unique endpoint you paste into your AI client, like .../services/mcp/v1/....
Custom role
A dedicated, low-permission role the AI runs as — never Administrator.
What to remember
- The AI is limited by the role behind it — same permissions as the NetSuite screens.
- Administrator is blocked on purpose — always use a dedicated, least-privilege role.
- It’s off by default — an admin must enable it and grant the MCP permission per role.
- Start read-only and small, watch the logs, then widen scope deliberately.
Quick summary: The NetSuite AI Connector Service is NetSuite’s implementation of an MCP (Model Context Protocol) server — a protocol-driven, “bring your own assistant” integration that lets AI clients such as Claude and ChatGPT query and act on NetSuite data through governed tools. The MCP Standard Tools SuiteApp ships pre-built tools for records, reports, saved searches, and read-only SuiteQL; you can also build custom tools in SuiteScript 2.1 and deploy them with SDF. Everything runs under a NetSuite role via OAuth 2.0, every call is logged, and the feature is disabled by default. This guide covers the architecture, prerequisites, permissions, OAuth, client connections, security, and the failure modes we actually hit in the field.
What is the NetSuite AI Connector Service?
Architecturally, the connector is an MCP server exposed on NetSuite’s SuiteTalk API domain. An AI client (the MCP “client”) connects to it, discovers the available tools, and invokes them in response to natural-language requests — translating “show me overdue invoices” into the right NetSuite operation and reading the result back. Because MCP is an open standard, NetSuite avoids vendor lock-in: any compliant client can connect to the same endpoints. (See NetSuite’s AI Connector Service docs.)
Two pieces do the heavy lifting. The MCP Standard Tools SuiteApp provides governed tools for working with records, reports, saved searches, and SuiteQL. And the connector enforces NetSuite’s existing role-based security — an AI client sees and does only what the connecting role permits, exactly as in the UI. (Reference: MCP Standard Tools SuiteApp.)
Prerequisites and licensing
On the NetSuite side you enable the feature and install the SuiteApp. The cost to plan for sits on the AI-client side:
- AI client plan — NetSuite documents a minimum of Claude Pro (or higher) or ChatGPT Plus (or higher) to add the connector. Other clients depend on their own MCP support.
- Account features — Server SuiteScript, REST Web Services, and OAuth 2.0 must be enabled (SuiteQL too if you use the query tools).
- The SuiteApp — install the MCP Standard Tools SuiteApp from the SuiteApp Marketplace (or build your own tools).
Required features and permissions
This is where most setups go wrong, so be precise. On each role that should use the connector, grant the right permissions — and note the easy-to-miss naming traps:
- MCP Server Connection — the permission that grants access to call MCP tools.
- Log in using OAuth 2.0 Access Tokens — not “Log in using Access Tokens.” The two are different; the wrong one fails silently.
- Record & search permissions — View or Edit on exactly the records the use case needs, plus Perform Search (often overlooked) for many standard tools.
- No Administrator — the connector refuses Administrator and full-permission roles by design. Build a dedicated role. (Reference: Required Features and Permissions.)
OAuth setup and integration records
Authentication is OAuth 2.0 Authorization Code Grant with PKCE; the client must also support remote MCP, streamable HTTP, and protocol version 2025-06-18. When you connect Claude or ChatGPT, NetSuite typically auto-creates the integration record on first connection — with Public Client checked, the Redirect URI populated, Authorization Code Grant enabled, and the AI Connector Service scope on.
Two behaviors catch teams out. An auto-created record can land in a pending state that an administrator must manually enable; once enabled it can be disabled but never returns to pending. And if a record exists without the AI Connector Service scope, you must recreate it. (Reference: AI Connector Service FAQ.)
Connecting Claude, ChatGPT, and other clients
Every client connects to a server URL that’s unique to your account. The two common forms:
# Only the MCP Standard Tools SuiteApp's tools
https://<accountid>.suitetalk.api.netsuite.com/services/mcp/v1/suiteapp/com.netsuite.mcpstandardtools
# All custom tools in the account (SuiteApps + ACPs)
https://<accountid>.suitetalk.api.netsuite.com/services/mcp/v1/all- Claude — in Settings → Connectors, add a custom connector with that URL, then complete the NetSuite OAuth prompt. Claude (Pro or higher) is the client Oracle documents most fully.
- ChatGPT — supported on Plus or higher; some tiers (e.g. Business) require enabling Developer Mode first. Important: as of March 4, 2026, ChatGPT moved to dynamic callback URLs, so a new integration record is needed for each new ChatGPT connection (pre-existing connections keep working).
- Gemini, Copilot, Cursor and others — these aren’t explicitly documented by Oracle. Any of them can connect only if it meets the four protocol requirements (remote MCP, streamable HTTP, OAuth 2.0 + PKCE, protocol 2025-06-18). Verify the client’s current MCP support against that checklist before promising it to stakeholders.
Security considerations
"Treat the AI Connector like any other integration: a dedicated least-privilege role, the exact scopes on the integration record matched in the role, and never an Administrator anywhere near it."
NetSuite builds in hard constraints — MCP tools never run as Administrator, can’t invoke elevated scripts or Suitelets, and can’t make external HTTP calls. Layer your own controls on top:
- Least privilege — one dedicated role per use case; never combine sensitive read access with broad write access.
- Explicit consent & logging — OAuth requires per-agent consent, and every MCP call is recorded in the integration record’s execution log (retention is around 21 days in production, 7 in sandbox).
- Trusted clients only — prompt injection and hallucination are inherent LLM risks NetSuite can’t eliminate; prefer clients that confirm before sensitive actions, and start with a limited scope.
Common implementation issues
- Empty tools list — usually the SuiteApp isn’t installed, or the role lacks MCP Server Connection / the right record permissions.
- OAuth failures — frequently the wrong permission (“Log in using Access Tokens” instead of the OAuth 2.0 variant), or an attempt to use Administrator.
- Integration record pending — an admin must enable it; if it was created without the AI Connector Service scope, recreate it.
- REST Web Services off — a classic oversight; the Standard Tools SuiteApp needs it, and calls fail without it.
- Large queries time out — AI clients cap around ~5,000 rows per call and have token limits; narrow the query.
Real-world use cases
The early wins are conversational reporting and lookups, not autonomous action:
- Finance — “What are our top 10 overdue invoices?” or “Has customer Z’s overdue balance been posted?” answered from live data.
- Sales — “Which pending deals are projected to close this month?” backed by a SuiteQL query the AI constructs.
- Support & ops — instant customer or item lookups without navigating NetSuite screens.
- Custom tools — package a SuiteScript 2.1 custom tool (e.g. a “deal risk” analysis) and let the AI invoke it by name. (Reference: Creating Custom Tools.)
Key takeaways
- The connector is NetSuite’s MCP server — bring your own AI client, governed by your roles.
- Plan licensing on the client side (Claude Pro / ChatGPT Plus or higher); enable Server SuiteScript, REST Web Services, OAuth 2.0.
- Permissions are the #1 trap — MCP Server Connection plus the OAuth 2.0 token permission, never Administrator.
- Connect via a unique server URL and OAuth; watch the ChatGPT dynamic-callback change from March 2026.
- Start read-only and small, log everything, and treat prompt injection as a real risk.
If you want this stood up safely on the first try, talk to one of our engineers — we’ll design the role, lock down the scope, and connect your AI client without exposing the account. You can also browse our other NetSuite engineering articles.
Frequently asked questions
What does the NetSuite AI Connector Service cost?
On the NetSuite side, the AI Connector Service is a protocol-driven feature you enable in your account, and the MCP Standard Tools SuiteApp is installed from the SuiteApp Marketplace. The cost you do need to plan for is on the AI client side: NetSuite documents that you need at least a Claude Pro plan or a ChatGPT Plus plan (or higher) to add the connector. Confirm current entitlements with your NetSuite account team, since AI offerings evolve quickly.
Which AI clients can connect — does it support Gemini?
The clients Oracle explicitly documents today are Anthropic’s Claude (Pro or higher) and OpenAI’s ChatGPT (Plus or higher). The service is built on the open Model Context Protocol, so any MCP-capable client that meets four requirements — remote MCP support, streamable HTTP, OAuth 2.0 Authorization Code Grant with PKCE, and protocol version 2025-06-18 — can connect. Gemini, Copilot and others are only usable if and when they meet those requirements, so verify the current capabilities of any non-documented client against that checklist.
Can the AI change my NetSuite data, or only read it?
It depends on the tools and the role. The MCP Standard Tools SuiteApp can read records, run reports, run saved searches and execute read-only SuiteQL, and it also includes tools that create and update records through REST Web Services. What any connection can actually do is bounded by the NetSuite role behind it — the AI runs with that role’s permissions and nothing more. For read-only assistants, grant a role with view-only access; never combine sensitive read access with broad write permissions.
Why can't I use the Administrator role with the AI Connector Service?
NetSuite deliberately blocks the Administrator role and other full-permission roles from using MCP. If an AI client or its credentials were ever compromised, an Administrator-scoped connection would expose the entire account, so the platform refuses it by design. The correct pattern is a dedicated, least-privilege custom role that holds only the MCP Server Connection and OAuth 2.0 permissions plus the specific record permissions the use case requires.
Is my NetSuite data sent to the AI vendor, and is it secure?
Data isn’t bulk-shared with the AI vendor; the AI client fetches only what a tool returns, on demand, through a protected OAuth-authenticated endpoint, scoped to the connecting role. Every MCP call is logged in the integration record’s execution log for auditability. The risks that remain are inherent to LLMs — prompt injection and hallucination — which NetSuite can’t eliminate, so use trusted clients, least-privilege roles, confirmation prompts for sensitive actions, and a limited initial scope.