How to use Claude with Oracle NetSuite
Short answer
Claude reaches NetSuite most cleanly through a Model Context Protocol connector that exposes a few read-only tools over SuiteQL or a RESTlet. You can run Claude through the Anthropic API, AWS Bedrock or Google Vertex AI, or keep it off the cloud and use an open-weight model with the same tools.
Finance and operations teams on NetSuite often want Claude for the same reasons they want any strong assistant: reading long documents, explaining a messy transaction trail, and drafting reconciliation notes. The difference with Claude is its support for MCP, which gives a standard way to hand it tools instead of building one-off plugins.
Oracle has been adding its own AI and connector capabilities to NetSuite, including support for MCP-style access in recent releases. Exact feature names, availability and licensing vary by account and release, so check what your edition includes before building your own.
This page lays out the practical setups, what to put on the NetSuite side, and when a hosted Claude is acceptable versus when you need data to stay under your control.
What teams use it for
Reconciliation notes from transaction history
Claude reads the linked transactions behind a mismatched intercompany balance and writes the explanation an accountant would otherwise piece together.
Contract and PO document review
Upload a supplier agreement from the File Cabinet and ask which clauses conflict with your standard terms, then link findings to the vendor record.
Searching NetSuite in plain English
A connector lets a user ask for overdue invoices by customer and receive a table built from a SuiteQL tool.
SuiteScript review and drafting
Developers use Claude to read an old User Event script and explain what it does before an upgrade or release change.
Audit evidence packs
For a sample of journal entries, Claude assembles the supporting records and a summary for the auditor request list.
Oracle NetSuite integration surface
- SuiteQL over REST Web Services supplies the read tools, and it handles joins that saved searches make awkward.
- RESTlets let you package a governed query or action as one tool with a stable contract.
- OAuth 2.0 or token-based authentication on an integration record, scoped to a dedicated role, protects the connector.
- The File Cabinet holds documents you may want to read, reachable through the record API or a RESTlet.
- Oracle's own MCP-style connector or AI features may be available in your release. Treat names and scope as release-dependent and verify them.
How it works
- 1
Check what your NetSuite release already offers
Review release notes and your account for any Oracle-provided connector or AI feature. Use it if it covers your need and your security team accepts it.
- 2
Choose the Claude route
Anthropic's API and Claude apps, AWS Bedrock and Google Vertex AI are the main hosted routes. Pick the one that matches your cloud contracts and data residency needs.
- 3
Build or adopt an MCP server
Expose read-only tools such as find_invoice, open_po_lines and get_transaction_chain. Give each a tight schema and a row limit.
- 4
Scope the NetSuite identity
Use a dedicated role that mirrors the least privileged user for the job, and avoid administrator tokens.
- 5
Connect and test with real questions
Run 30 to 50 real questions from users, check each answer against NetSuite, and record failures caused by custom fields or segments.
- 6
Decide on writes later
Only after read quality is proven, add draft-creating tools with approval in NetSuite.
Example flows
Explain an intercompany difference
Trigger: An accountant asks why two subsidiaries disagree on an intercompany balance.
- Claude calls a tool that returns intercompany transactions for both subsidiaries.
- It lines up matching pairs and flags the unmatched ones.
- It identifies timing, currency or missing elimination as the cause.
- It writes a note for the accountant to review.
Result: A multi-hour reconciliation becomes a reviewed explanation with transaction IDs.
Review a supplier agreement
Trigger: Procurement uploads a new supplier contract against the vendor record.
- Claude reads the document and your standard terms checklist.
- It lists deviations in payment terms, liability and delivery.
- It posts the summary as a note on the vendor for the buyer.
Result: The buyer enters negotiation knowing which clauses to challenge.
Keeping data on-prem
NetSuite runs in Oracle's cloud, so on-prem here means where the model and the MCP server run, not the ERP. If data rules rule out any hosted Claude route, swap in an open-weight model on your own GPUs behind the same MCP server and keep tools, prompts and logs inside your network. The MCP layer is the reusable part, so you can start with hosted Claude on low-risk questions and move sensitive workflows to the private model later without rebuilding the NetSuite side. Where a hosted route is acceptable, Bedrock or Vertex can help you keep traffic within an existing cloud boundary and region.
Security checklist
- Run the MCP server with a dedicated NetSuite role and a minimum set of tools.
- Use HTTP transport with authentication for shared servers, and reserve stdio for a single developer's local use.
- Return only approved fields and cap rows per call.
- Log every tool call with the user and parameters for audit.
- Review the data handling terms of your chosen Claude route against your contracts and classifications.
Ways to get it done
Use Oracle's own connector or AI features if present
Effort: Configuration and security review.
Best when your release includes what you need and licensing is acceptable.
Build an MCP server in-house
Effort: Several weeks for a solid read-only version.
Good for teams that want custom tools over SuiteQL and RESTlets.
Netray-built connector with a hosted or private model
Effort: A scoped pilot, then extend.
Useful when you want evaluation, masking and an on-prem path included.
Pitfalls
- !Connecting Claude with an administrator token because it was quick to set up.
- !Exposing a free-form SQL tool, which lets the model run heavy or sensitive queries.
- !Not describing custom fields and segments, so Claude guesses what they mean.
- !Treating a pilot demo as proof, without a replay of real user questions.
- !Ignoring subsidiary restrictions in OneWorld accounts.
Frequently asked questions
Does Claude have a built-in NetSuite connector?
Availability changes, and both Oracle and Anthropic have been adding connector options. Check Claude's current connector directory and your NetSuite release notes. If nothing fits, a small MCP server over SuiteQL or a RESTlet is a dependable route that you control.
Which is better, Anthropic API, Bedrock or Vertex?
They offer the same family of models with different contracts, regions and billing. Pick based on where your security and procurement teams already have agreements. Bedrock suits AWS-centric shops, Vertex suits Google Cloud users, and the direct API is simplest to start with.
Can Claude change data in NetSuite?
It can if you expose write tools, but start with read-only. When you add writes, have Claude create drafts or pending records and require approval by a NetSuite user before anything posts.
Is a hosted Claude acceptable for financial data?
Often yes, under enterprise terms and with the right controls, but it depends on your contracts, regulators and data classification. Export-controlled or customer-restricted data usually calls for a private model hosted by you.
What is MCP and why use it here?
Model Context Protocol is an open standard for giving AI clients tools and data. Build the NetSuite tools once and they work with Claude, ChatGPT and other MCP clients, which keeps you from rewriting integrations when you change models.
Want Claude working on your Oracle NetSuite?
A free 30 minute assessment: we map the integration path for your version and setup, and where it should run.
Get a free assessment