What can Claude do with NetSuite? See the use cases
To connect Claude to NetSuite, enable Server SuiteScript, OAuth 2.0, and REST Web Services; create a non-administrator role with the MCP Server Connection and Log in using OAuth 2.0 Access Tokens permissions; install the free MCP Standard Tools SuiteApp; then add the NetSuite connector in Claude and authorize it while signed in with that role.
| Attribute | Answer |
|---|---|
| Product | NetSuite AI Connector Service, from Oracle NetSuite |
| Protocol | Remote Model Context Protocol (MCP) over streamable HTTP, with OAuth 2.0 and PKCE |
| Oracle cost | Not a paid NetSuite feature; the MCP Standard Tools SuiteApp is free. Your AI client may need a paid plan. |
| Administrator role | Not supported. Use a custom, non-administrator role. |
| Standard tools | Records, reports, saved searches, and read-only SuiteQL |
| Audit trail | Execution Log on the integration record, retained 21 days in production and 7 in sandbox |
This guide covers both halves of connecting Claude to NetSuite: the setup steps for Oracle's NetSuite AI Connector Service, and the role, permission, and review design that keeps the connection safe six months later. The setup half follows Oracle's documentation, linked throughout. The governance half is our approach: read access before write access, and a person approving anything that posts.
The NetSuite AI Connector Service is Oracle's implementation of the Model Context Protocol for NetSuite. It lets an external AI client, such as Claude, call tools that work with your NetSuite data. Oracle provides the MCP Standard Tools SuiteApp with ready-made tools for records, reports, saved searches, and SuiteQL, and developers can build custom tools in SuiteScript (Oracle: NetSuite AI Connector Service).
Two properties matter most for finance teams. Every tool runs with the permissions of the NetSuite role you connect with, so the AI cannot see or do anything that role cannot. And Oracle states that NetSuite cannot control how an external AI client handles data once it leaves NetSuite, so the choice of client and the data you expose are your governance decisions (Oracle: AI Connector Service FAQ).
This page covers setup and governance. For what finance teams do once Claude is connected, read what a Claude and NetSuite integration can do.
Oracle documents two paths in Claude: the listed NetSuite connector for the standard tools, and Add custom connector under Settings > Connectors for custom tool URLs. The steps below follow the standard tools path (Oracle: Connect to the AI Connector Service). Anthropic's own guidance on remote MCP connectors is in the Claude Help Center.
The connector can only do what the role can do, which makes role design the security model. The default failure is building one broad role because it is fastest and everything works. That gives an AI client the ability to touch anything the role touches, and makes it hard to tell from an audit trail whether a change came from a person or a process.
Note that the MCP Standard Tools SuiteApp includes tools that create and update records through REST Web Services, and any role with MCP Server Connection can use them on any record its permissions allow. Oracle recommends a separate role holding only the record permissions users should have through an AI client. Our design rules:
Post to a closed period. Change a posting period status. Modify an item's costing method. Approve anything it also created. Alter a customer's credit terms. Touch bank reconciliation. These are not model capability limits, they are control environment limits, and they should be enforced by permissions rather than by instructions in a prompt.
The highest-value early workflows are ones where the agent assembles and a person commits. An A/R aging with exceptions flagged and payment matching attempted. A reconciliation with the variances isolated and explained. A close checklist with the open items identified. The agent does the searching and assembling, which is where the hours go, and a person makes the decision, which is where the risk is.
The standard tools retrieve data three ways: SuiteQL queries, saved searches, and reports. Oracle notes that the SuiteQL tools are read-only, that report tools accept only date and subsidiary filters (not accounting periods), and that AI clients typically handle up to roughly 5,000 rows per call, with very large queries at risk of timing out.
In practice, use saved searches for anything finance already trusts, because the logic is reviewed and reused. Use SuiteQL for exploratory questions and narrow lookups, and ask Claude to show the query it ran so a person can check it. Avoid pulling wide record sets and asking the model to summarize them: it is slower, costs more, and is more error-prone than a well-formed query that returns exactly what is needed. Oracle also cautions that AI output can be wrong, so verify results against NetSuite before they are relied on.
| Symptom | What to check |
|---|---|
| Cannot connect after adding permissions | Permissions are on the role you signed in with; it is Log in using OAuth 2.0 Access Tokens, not Log in using Access Tokens; the role is not Administrator; required features are enabled |
| Connected, but no tools appear | The MCP Standard Tools SuiteApp is installed, the role has the required permissions, and access to the SuiteApp folder is not restricted |
| Claude shows the connector as disconnected | The server URL is correct, including /all when calling all tools, and the SuiteApp is installed |
| Too Many Requests errors | Your account concurrency limit, which is shared with other integrations; administrators can allocate limits per integration record |
| Integration record stuck as pending | Enable it under Setup > Integration > Manage Integrations; once enabled it can be disabled but not returned to pending |
| Report numbers look wrong for a period | Report tools filter by date and subsidiary only; use a saved search or SuiteQL for period-based questions |
Each request appears in the integration record's Execution Log with time, duration, status, user, HTTP response, and request details. Oracle retains these entries for 21 days in production and 7 days in sandbox, so export or review them on a schedule if your audit trail needs to be longer. If your team needs help with role design, monitoring, or custom tools, see MCP integration consulting and SuiteScript development.
A purpose-built, non-administrator role scoped to the workflow. Oracle does not support the Administrator role for the NetSuite AI Connector Service. Grant only the record permissions the workflow needs, keep posting permissions off unless a workflow requires them with an approval step, and assign the role to named users so activity is attributable.
No. Tools run with the permissions of the role you connect with. Whatever a tool returns is sent to the AI client, though, and Oracle notes it cannot control how external clients handle data once it leaves NetSuite, so scope the role to data you are comfortable sharing.
The standard tools can create and update records through REST Web Services if the role allows it. Whether it should is a control question. Our default is that Claude proposes and a person posts, with write permissions added per workflow rather than granted at setup.
Not if each connection runs under a named user with a purpose-built role. Requests are recorded in the integration record's Execution Log. It breaks if people share credentials, because then you cannot tell whose session made a change.
Oracle states it is not a paid NetSuite feature and that the MCP Standard Tools SuiteApp is free. Your AI client may require a paid plan.
Oracle states the service has not been assessed for HIPAA compliance and should not be used with electronic protected health information unless you have independently determined that use meets your obligations.
Oracle lists prompt injection, hallucination, tool poisoning, and tool drift, along with compromised clients or servers. Mitigations include trusted clients only, least-privilege roles, limiting which tools are available, regular access reviews, and monitoring the Execution Log.
Not to get started. The standard tools cover records, reports, saved searches, and SuiteQL. Custom tools written in SuiteScript 2.1 make sense for specialized or single-step workflows; see SuiteScript development.
Beyond your AI client plan, it depends almost entirely on how data retrieval is designed. Poorly scoped retrieval is where runaway cost comes from, and it is straightforward to avoid at design time and painful to fix later.
Work with us
Every Claude engagement starts with the AI Action Plan. It is the only fixed price we publish, and it tells you which workflows deserve Claude at all.
$5,000 fixed fee
Two weeks from kickoff. Classifies your finance and operations workflows and sequences what is worth building, before any AI spend.
Scoped after the Action Plan
Workspace setup, reusable instructions and skills, and governance boundaries for your accounting team.
Scoped after the Action Plan
Purpose-built roles, read-only first workflows, and Execution Log monitoring.
TFR Solutions designs roles, permissions, guardrails, and telemetry for AI agents connected to ERP and operational
Explore AI for finance teams with human checkpoints, observable reasoning, and read-before-write architecture designed
Explore five AI architecture levels, their requirements, and how mid-market companies can choose an appropriate
The $5,000 fixed-fee AI Action Plan classifies workflows, identifies AI candidates, and delivers a sequenced backlog
30 minutes. No commitment. We will tell you honestly whether we can help.
Start Your AI Action PlanExplore AI ServicesLast updated:
By Teddie Reyes ยท Last reviewed: August 2026
Claude is a trademark of Anthropic, PBC. TFR Solutions is not affiliated with Anthropic, PBC.
TFR Solutions, Los Angeles, CA · info@tfr.solutions · (310) 894-6031