Contents10 sections ↓
What is NetSuite MCP?
NetSuite MCP is Oracle's implementation of the Model Context Protocol, the open standard for connecting AI assistants to external systems. Oracle ships it as the NetSuite AI Connector Service: an MCP server inside your NetSuite account that external assistants such as Claude and ChatGPT can call, plus the MCP Standard Tools SuiteApp, which supplies the tools those assistants use.
Direct answer: NetSuite MCP is the NetSuite AI Connector Service, Oracle's MCP server for NetSuite. Install the MCP Standard Tools SuiteApp, enable Server SuiteScript and OAuth 2.0, and give a non-administrator role the "MCP Server Connection" permission. Each user then authorizes Claude, ChatGPT or another MCP client with OAuth 2.0, and the assistant can query, report on and update NetSuite within that role's permissions.
Oracle positions it as "bring your own AI": you choose the assistant, NetSuite controls what it can see and do. This post explains how it works, what it can and can't do, and how it is secured. For a click-by-click Claude setup, see How to connect NetSuite to Claude. For where MCP fits among NetSuite's other AI features, see the NetSuite AI guide.
Checked against Oracle's help center on 21 October 2026.
How NetSuite MCP works
- Your AI client (Claude, ChatGPT, or another MCP client) receives a request in natural language.
- The client calls NetSuite's MCP server at an account-specific URL, after the user has authorized it with OAuth 2.0.
- NetSuite runs the MCP tool under that user's role and permissions.
- Results return to the client, which summarizes, charts or acts on them.
The assistant never gets database access. Everything goes through tools, and every tool runs with the same permissions the user has in NetSuite.
Server URLs
From Oracle's connection guide:
| URL | Returns |
|---|---|
https://<accountid>.suitetalk.api.netsuite.com/services/mcp/v1/suiteapp/com.netsuite.mcpstandardtools | The tools in the MCP Standard Tools SuiteApp |
https://<accountid>.suitetalk.api.netsuite.com/services/mcp/v1/suiteapp/<applicationid> | Only the custom tools exposed by one SuiteApp (application ID is <publisherid>.<projectid>) |
https://<accountid>.suitetalk.api.netsuite.com/services/mcp/v1/all | All custom tools in the account, from SuiteApps and account customization projects |
Pointing a client at a single SuiteApp URL is also how users scope an assistant to a narrower set of tools.
The MCP Standard Tools SuiteApp
The MCP Standard Tools SuiteApp is installed from the SuiteApp Marketplace. Oracle's tool list has 25 tools in seven groups:
| Group | Tools | What they do |
|---|---|---|
| Record | ns_createRecord, ns_getRecord, ns_updateRecord, ns_transformRecord, ns_getRecordTypeMetadata, ns_getRecordTransformMetadata | Create, read, update and transform records (for example sales order to invoice); read field metadata |
| Record Action | ns_getRecordActionMetadata, ns_executeRecordAction | Discover and run actions available on a record |
| Report | ns_listAllReports, ns_runReport, ns_getSubsidiaries, ns_getAccountingBooks, ns_getAccountingContexts, ns_getNexusIds | List and run standard and custom reports, with filter values |
| Saved Search | ns_listSavedSearches, ns_runSavedSearch | List and run existing saved searches |
| SuiteQL | ns_runCustomSuiteQL, ns_getSuiteQLMetadata | Run read-only SuiteQL; read queryable fields and joins |
| File Cabinet | ns_attachFile, ns_detachFile, ns_getFolderByPath, ns_getIdFromPath, ns_getPathFromId, ns_searchFiles | Browse, search, attach and detach files |
| Runtime | ns_getSessionInfo | Return account, role, enabled features and permissions for the session |
A tool only appears in the AI client if the user's role has every permission it requires, and visibility does not widen access: the role's permissions still decide what data comes back. Custom SuiteQL is read-only by design.
The SuiteApp also ships MCP Apps, interactive interfaces that compatible AI clients render inside the chat: a File Cabinet browser, file get and upload, a record selector, report filters and a prompt library (Oracle's list).
A second SuiteApp, the NetSuite AI Connector Service Companion, adds what Oracle describes as "over 100 prompt templates", preconfigured "MCP-Ready Roles" (CFO, Controller, AR/AP Analyst, Treasury) and Skills for supported AI models (product page).
If the standard tools don't fit, developers can create custom tools in SuiteScript and deploy them in a SuiteApp or account customization project.
Requirements and setup overview
From Oracle's Required Features and Permissions:
| Requirement | Where |
|---|---|
| Server SuiteScript feature | Setup > Company > Enable Features > SuiteCloud |
| OAuth 2.0 feature | Same subtab |
| MCP Server Connection permission | Setup > Users/Roles > Manage Roles > Permissions > Setup |
| Log in using OAuth 2.0 Access Tokens permission (not "Log in using Access Tokens") | Same subtab |
| REST Web Services feature and permission | Needed for the Standard Tools to create, read and update records |
Oracle does not allow the connector to run under the Administrator role or any role with full permissions. Create a dedicated role per business function with only the record permissions it needs; Oracle notes that a role with MCP Server Connection can create and update any record available through REST if its other permissions allow it.
Authentication is OAuth 2.0 authorization code grant with a public client. For Claude, the integration record (with the NetSuite AI Connector Service scope) is installed automatically on the first connection; for other clients an administrator may need to create or enable it. Every user consents to every AI agent during the OAuth flow.
Which AI clients work with NetSuite MCP?
| Client | What Oracle documents |
|---|---|
| Claude (claude.ai) | NetSuite connector under the Web tab for the Standard Tools; custom connector for custom tools. Oracle says Claude needs "the Pro plan (or higher)" |
| ChatGPT | Connect through Apps, search for NetSuite, sign in and pick a non-administrator role. Custom tools need a new integration record per connection |
| Claude Code, Codex, Copilot Studio, Postman | Covered in SuiteAnswers articles linked from Oracle's connection guide |
| Other MCP clients | Oracle says to "adjust the instructions accordingly" |
The full Claude walk-through is in How to connect NetSuite to Claude.
Concurrency
Unless an administrator sets a concurrency limit on its integration record, the connector shares your account's concurrent-request limit with every other integration. When the limit is hit the client gets a "Too Many Requests" error and must retry. Each tool call is also preceded by MCP protocol requests, which count too (Oracle). If you run heavy integrations, give the connector its own allocation.
Want NetSuite MCP set up with least-privilege roles?
We configure the AI Connector Service, build dedicated roles per team, and write custom MCP tools for your own records and processes.
Talk to a NetSuite AI developerWhat MCP can and can't do in NetSuite
| MCP can | MCP can't |
|---|---|
| Read records and record metadata | Run under Administrator or full-permission roles |
| Run standard and custom reports and saved searches | Use "Run as role" or invoke scripts with elevated privileges |
| Run read-only SuiteQL | Write data through SuiteQL |
| Create, update and transform records, if the role allows | Invoke Suitelets |
| Run record actions | Make HTTP requests to external destinations |
| Attach files from the File Cabinet | See anything the role can't see |
| Call custom tools you build | Use NetSuite AI Units (the connector doesn't consume them) |
What it is good at in practice: ad hoc questions a saved search was never built for ("which customers over 60 days overdue also have open sales orders?"), pulling a report into a conversation for analysis, drafting records from an email or document the user pastes, and letting developers check data while they work.
What it is not: a replacement for system-to-system integration. A nightly order sync from Shopify, a 3PL feed or a bank import should stay on REST web services, RESTlets or an iPaaS, where the logic is fixed, tested and monitored. MCP is for a person working with an assistant. Our NetSuite REST API guide covers the integration side.
Two accuracy notes from Oracle's best practices: have the model use ns_runReport or ns_runSavedSearch before custom SuiteQL, and do not rebuild financial reports with SuiteQL, because standard reports apply business rules SuiteQL can't. When creating records, set an External ID to avoid duplicates.
MCP vs. other ways to integrate
| Aspect | MCP (AI Connector Service) | REST / SOAP web services | RESTlets |
|---|---|---|---|
| Caller | An AI assistant acting for a user | Another system | Another system |
| Custom code | None for the standard tools | Yours, in the calling system | SuiteScript |
| Queries | Chosen by the model at run time | Fixed by your code | Fixed by your code |
| Auth | OAuth 2.0 per user, per agent | OAuth 2.0 (TBA for existing integrations) | OAuth 2.0 (TBA for existing integrations) |
| Best for | Ad hoc questions, analysis, assisted record work | Scheduled, high-volume system integration | Custom business logic endpoints |
On authentication: from NetSuite 2027.1, new token-based authentication integrations can't be created, NLAuth for RESTlets ends, and new OAuth 2.0 authorization-code integrations must use PKCE. Full TBA end of support is tentatively 2028.2. MCP already runs on OAuth 2.0.
NetSuite MCP vs. the N/llm module
MCP lets external assistants reach into NetSuite. The N/llm module lets your server scripts inside NetSuite call an LLM that Oracle provides. They point in opposite directions.
| MCP | N/llm module | |
|---|---|---|
| Direction | External AI client into NetSuite | NetSuite script out to an LLM |
| Who uses it | People chatting with Claude, ChatGPT and others | Developers building automations |
| Model | Yours | Oracle-provided |
| Where logic lives | The conversation | Your SuiteScript |
| AI Units | Not consumed | Consumed per call |
| Best for | Ad hoc questions, analysis, assisted workflows | Embedded text generation, classification, tool-calling automations |
They compose well: an MCP-connected assistant for your team's questions, and N/llm automations for repetitive text work inside processes.
NetSuite MCP vs. Ask Oracle
| NetSuite MCP | Ask Oracle | |
|---|---|---|
| AI model | Bring your own | Oracle's built-in models |
| Interface | Your AI client | Inside NetSuite (NetSuite Next) |
| Availability | Separate from NetSuite Next; an administrator enables it | Rolling out with 2026.2 in the US and Canada |
| Data access | What the role permits | What the role permits |
| AI Units | Not consumed | Consumed |
| Cost beyond NetSuite | Your AI client's subscription | None beyond AI Units |
Oracle's own Ask Oracle page says it "only interacts with your data within NetSuite" and points to the AI Connector Service for external LLMs. Many teams will use both. See the NetSuite Next guide for Ask Oracle.
Security and governance
Oracle's risks, controls and mitigation strategies page is the document to read before enabling MCP.
Controls in NetSuite
- MCP is off by default; no user has access until an administrator grants the permission to a role.
- Tools run with the permissions of the user behind the AI agent, never with Administrator or full-permission roles.
- "Run as role" is disabled for all tools; tools can't invoke elevated scripts or Suitelets or make external HTTP requests.
- All MCP tool usage is logged.
- During OAuth, the MCP server obtains explicit consent from each user for every AI agent.
- Users can scope tools for an agent by namespace.
Risks Oracle names
- Prompt injection: hidden instructions in PDFs, web pages or third-party tool responses that make the agent act or leak data.
- Hallucination: plausible but wrong answers.
- Tool poisoning: a malicious tool hiding instructions in its description or schema.
- Rug pull or tool drift: a tool changing behavior after it was approved.
- Compromised external agents or servers, and clients that don't enforce NetSuite's policies.
Oracle states that prompt injection and hallucination are "LLM weaknesses and out of NetSuite control".
Practices we follow
- Grant MCP Server Connection only to dedicated roles, one per business function, with the minimum record permissions.
- Start read-only: leave REST Web Services write permissions off until the team trusts the answers.
- Use clients that ask for confirmation before creating, changing, approving or paying.
- Don't combine NetSuite's MCP tools in the same agent with tools that have broad access to email, browsers or local files.
- Keep an approved list of AI clients and MCP servers, and review roles, consents and logs on a schedule.
- Document how to revoke access quickly (remove the permission, revoke consent, disable the integration record).
One compliance note: Oracle states the AI Connector Service has not been assessed for HIPAA and must not be used for ePHI unless you have independently determined that use is compliant.
What we've learned building AI for NetSuite
We were connecting AI to NetSuite before MCP existed. We built ContextQL, a tool that teaches AI NetSuite's data model and writes SuiteQL. That shapes how we approach MCP:
Start with read-only. Let the assistant query before it writes.
Define use cases first. "Connect AI to NetSuite" is not a use case. "Weekly AR aging analysis for the controller" is.
Describe your custom objects. The standard tools read metadata, so field and record descriptions decide how well the model understands custom records. Fill in AI descriptions on custom records and fields.
Test with real volumes. A query that works on ten records may time out or hit concurrency limits on ten thousand.
Keep humans in the loop. AI is good at finding patterns. Approving purchase orders and posting journals stays with people.
What clients ask before signing
Get your integration scoped
Tell us what systems you need to connect. We'll recommend the right approach and ballpark the cost.

Sebastian Correa
Co-Founder & CCO
Co-founder and Chief Creative Officer at BrokenRubik with 12+ years of experience in NetSuite consulting and e-commerce development. Specializes in helping businesses optimize their ERP operations and scale their online presence through strategic technology implementations.
