# How to use Claude with Priority ERP

> Source: https://netray.co/integrations/priority-claude
> Publisher: Netray (netray.co), AI and integration engineering for ERP and manufacturing systems.
> Section: Integrations

## Short answer

Claude reaches Priority most cleanly through a Model Context Protocol server that wraps the Priority OData API with read-only tools. You can call Claude through Anthropic's API, AWS Bedrock or Google Vertex, and keep an open-weight model on-prem for data that cannot leave.

## System

Priority ERP

## Capability

claude

## Intro

Teams on Priority who try Claude usually start by pasting reports into a chat window. It is a good way to see what the model can do with your real wording, such as part descriptions, order notes and supplier remarks, which are often messy and multilingual.

To make it dependable, move from pasting to a connector. A connector gives Claude a defined set of tools, each backed by a Priority OData query, so answers come from current records and cite document numbers. The connector is also where you enforce which entities, fields and companies are visible.

Hosting is the second decision. Claude can be reached directly through Anthropic's API, through AWS Bedrock or Google Vertex AI inside a cloud account you already govern, or not at all for the most restricted data, where a local open-weight model takes over.

## Use Cases

### Reading long supplier and customer correspondence

**Description:** Claude summarizes email threads attached to a purchase order or sales order and extracts commitments, dates and open questions for the buyer.

### Explaining a BOM change

**Description:** Give Claude the old and new BOM levels and ask what changed and which work orders are affected. Useful for engineering change reviews.

### Quote and RFQ drafting

**Description:** Using part, cost and lead time lookups from Priority, Claude drafts a quote response in the customer's wording for a salesperson to edit.

### Analyst-style questions over live data

**Description:** Through connector tools, a manager asks which customers have orders slipping past promise date and why, and gets an answer with the order numbers.

### Documenting custom logic

**Description:** Developers use Claude to explain existing procedures and report code and to draft documentation. Output is checked against the actual system.

## System Surface

- Priority OData v4 over REST is the main data path. A connector maps tools like get_order, list_open_pos and get_part_stock to specific entity queries.
- Priority forms and sub-forms form a hierarchy, so a tool can return an order with its lines in one call using expansion where your version supports it.
- Attachments and remarks are usually exposed as sub-form fields or linked files. Decide explicitly whether Claude may read them, since they carry free text.
- Service users and application credentials control access. The connector should authenticate with a Priority user whose profile is limited to the forms the tools need.
- Exact entity names, expansion support and header conventions differ by version and deployment, so verify in a test company first.

## How It Works

### Decide which questions Claude should answer

**Detail:** Write ten to twenty real questions your team asks. They define the tools you need and become the test set.

### Build an MCP server over Priority OData

**Detail:** Expose each question type as a read-only tool with typed parameters. Add limits, field whitelists and a company filter. Keep Priority credentials in the server.

### Choose the transport

**Detail:** Use stdio for a developer on a workstation, and HTTP with authentication for a shared service that several users reach.

### Pick the Claude access path

**Detail:** Use a Claude plan that fits your data policy, or call models via the Anthropic API, AWS Bedrock or Vertex AI so traffic stays in an account you already control.

### Evaluate against the test set

**Detail:** Check correctness, citations and refusals. Fix tool descriptions before blaming the model, because unclear descriptions cause most wrong calls.

### Roll out by group

**Detail:** Start with planners or customer service, review the logs, and widen access only when the answers are consistently sound.

## Example Flows

### Late order explanation

**Trigger:** Customer service asks why order is late

**Steps:**

- Claude calls get_order for the order header and lines.
- It calls get_work_order_status for the linked production.
- It calls list_open_pos for the missing component.
- It explains the chain with document numbers and the new expected date.

**Result:** A clear reply the service rep can send, with every claim traceable.

### RFQ response draft

**Trigger:** A salesperson pastes a customer RFQ

**Steps:**

- Claude extracts requested parts and quantities.
- Tools return standard cost, stock and typical lead time.
- Claude drafts the reply and flags parts it could not find.

**Result:** A near-ready draft, with unknown parts clearly marked for human follow-up.

## On Prem

If your Priority data is controlled technical data or customer-restricted, a hosted model may not be allowed. In that case, keep the same connector design and swap the model for an open-weight one served locally with vLLM or Ollama, reading Priority over the LAN. Some customers run a split: Claude via a governed cloud account for general business questions, and a local model for restricted programs. The MCP server is the shared piece, so the tools and permissions stay the same either way.

## Security

- Review the data handling terms of the specific Claude plan or cloud platform you choose, including retention and training exclusions.
- Run the connector with a read-only Priority user and a field whitelist, and enforce limits in code, not in the prompt.
- Treat free-text fields as untrusted input, since remarks or emails could contain instructions aimed at the model.
- Authenticate every HTTP connector request and tie it to a named user for audit.

## Options

### Pasted exports into a Claude workspace

**Fit:** Fine for non-sensitive analysis and drafting. No live data.

**Effort:** None

### In-house MCP connector plus Claude via API, Bedrock or Vertex

**Fit:** Good if you have a developer and a cloud account with an approved data policy.

**Effort:** Two to four weeks for a read-only set

### Netray-built connector with an on-prem fallback model

**Fit:** Suited to regulated manufacturers who want one tool layer and a choice of where the model runs.

**Effort:** Scoped project

## Pitfalls

- Writing vague tool descriptions, which leads Claude to choose the wrong tool or invent parameters.
- Returning thousands of rows to the model instead of aggregating in the connector.
- Mixing companies or legal entities in one session without an explicit filter.
- Assuming a cloud platform's regional setting covers every subprocessor, without checking.

## Faq

### Does Priority have a native Claude integration?

**Answer:** I am not aware of a standard one, and this changes quickly. Ask your Priority partner and check the Claude connector directory. Otherwise an MCP server over the Priority OData API is the dependable route.

### Can Claude run entirely inside our network?

**Answer:** Not as a self-hosted model you control. If you need that, use an open-weight model on your own hardware with the same connector. If a governed cloud account is acceptable, Bedrock or Vertex can keep traffic inside your cloud boundary.

### Will Claude change records in Priority?

**Answer:** Not unless you give it write tools. Start read-only. If you add writes, make them create drafts that a user approves, and use a Priority user with minimal rights.

### How accurate are the answers?

**Answer:** Accuracy depends on the tools, the data and your master data quality. Measure it on your own question set and require citations of document numbers, so users can verify.

### Which Claude access path should we use?

**Answer:** Pick by data policy and procurement. A direct plan is quickest to try. Bedrock or Vertex suits teams that already govern a cloud account. For restricted data, use a local model instead.

## Search terms

priority erp claude, claude ai priority erp, connect claude to priority erp, claude mcp priority software, priority erp anthropic, claude bedrock priority erp, claude for erp manufacturing, priority erp llm integration, claude on-prem alternative erp, use claude with priority

---
Talk to an engineer: https://netray.co/contact?intent=erp_ai | Machine catalog: https://netray.co/agent/catalog.json