How to Set Up the NetSuite AI Connector Service (MCP)
A practical, step-by-step guide that turns the playbook into a working setup — enable the features, install the MCP Standard Tools SuiteApp, build a least-privilege role, configure OAuth, connect Claude or ChatGPT, then lock it down — with links to the official NetSuite documentation so you can verify everything for yourself.
Practical developer guide · Backed by the official NetSuite Help Center
This is the practical companion to our NetSuite AI Connector Service guide. The main article explains what the connector is and how it’s governed; this page is the plain checklist for standing it up. Work through the steps in order — least-privilege role, sandbox-first — and use the official NetSuite documentation link under every step to confirm the exact feature, permission, and URL as you go.
Before you start — prerequisites
A sandbox: stand the connector up in a Sandbox or Release Preview account first, and validate before touching Production.
An AI client plan: Claude Pro (or higher) or ChatGPT Plus (or higher); other MCP-capable clients depend on their own support.
Admin access & a security sign-off: you’ll enable features and create a role — agree the data scope with whoever owns security first.
What you'll set up
Enable the required NetSuite features
Install the MCP Standard Tools SuiteApp
Create a dedicated least-privilege role
Configure OAuth 2.0 and the integration record
Get the server URL and connect your AI client
Lock down security and governance
Test with natural-language queries and roll out
1
Enable the required NetSuite features
The connector won’t work until the underlying platform features are on. Missing REST Web Services is the single most common reason calls fail.
Go to: Setup > Company > Enable Features
Under SuiteCloud, enable Server SuiteScript, REST Web Services, and OAuth 2.0.
Enable SuiteQL/SuiteAnalytics features if you intend to use the query tools.
Save and confirm the features are active before moving on.
This SuiteApp registers the pre-built tools an AI client will discover — records, reports, saved searches, and SuiteQL. Without tools, the connection has nothing to call.
Go to: SuiteApps (Marketplace) → install MCP Standard Tools
Install the MCP Standard Tools SuiteApp from the SuiteApp Marketplace.
Confirm the tools appear under Customization > Scripting > Custom Tools.
Note the SuiteApp’s application ID — you’ll need it for the server URL in Step 5.
Authentication is OAuth 2.0 (Authorization Code Grant with PKCE). When you connect Claude or ChatGPT, NetSuite usually auto-creates the integration record — but you must confirm it’s enabled and correctly scoped.
Go to: Setup > Integration > Manage Integrations
Connect once from the AI client to trigger the integration record, then open it under Manage Integrations.
Confirm it’s enabled (not pending) with the AI Connector Service scope, Public Client, and Authorization Code Grant set.
If the scope is missing or the record won’t enable, delete and recreate it following the OAuth 2.0 guidance.
Use .../services/mcp/v1/suiteapp/com.netsuite.mcpstandardtools for the standard tools, or .../services/mcp/v1/all for all custom tools.
Claude: Settings → Connectors → add a custom connector with the URL, then complete the NetSuite OAuth prompt.
ChatGPT: Plus or higher; enable Developer Mode where required. Since March 4, 2026, dynamic callback URLs mean a new integration record per new connection.
Prove it with a simple question, confirm the data and the log entry match, then widen scope deliberately rather than all at once.
Test: “How many customers do we have in NetSuite?”
Ask a simple read query in the AI client and confirm the number matches live NetSuite data.
Check the execution log shows the call (method, URL path, user, status) for traceability.
Widen scope slowly — add tools and permissions only as each use case is validated. For bespoke needs, build a custom tool in SuiteScript 2.1 and deploy it via SDF.
A note on accuracy: the AI Connector Service is evolving quickly — client requirements, supported plans, and even callback behavior change between releases (for example, ChatGPT’s move to dynamic callback URLs on March 4, 2026). Always confirm each feature, permission, and URL against its official reference, and validate in Sandbox before enabling anyone in Production.
Everything above gets a governed AI client reading live NetSuite data safely. What sits beyond the standard tools — custom tools in SuiteScript 2.1 with JSON-schema definitions, interactive MCP Apps that render a bundled UI inside the AI client, role designs that pass segregation-of-duties testing, and change-control on the integration record and toolsets — is real engineering work. It usually means SDF-deployed custom tools and a governance policy, not just a URL pasted into a chat client.
This feature points an external LLM at your ERP, so build it in a Sandbox, keep the role least-privilege, watch the execution log, and widen scope only as each use case proves out. For the full reasoning — the architecture, licensing, security model, and failure modes — read the companion guide.
Connecting AI to NetSuite?Our NetSuite engineers design the least-privilege role, configure OAuth and the MCP tools, connect Claude or ChatGPT, and set the governance your security team needs — Sandbox-first, so nothing leaks.
Documentation links point to the official Oracle NetSuite Help Center (docs.oracle.com). The AI Connector Service evolves rapidly — exact menu paths, supported AI-client plans, protocol requirements, and connection behavior vary by NetSuite version and by client; confirm each step against the linked Help Center page and validate in Sandbox first. Claude is a product of Anthropic; ChatGPT of OpenAI — check each provider’s own documentation for current plan requirements.