# How to use Claude with IFS Cloud

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

## Short answer

Use Claude with IFS Cloud by exposing IFS OData data through an MCP server that Claude can call, or by calling Claude from your own integration through the Anthropic API, AWS Bedrock or Google Vertex AI. For restricted data, pair the same tools with a privately hosted model.

## System

IFS Cloud

## Capability

claude

## Intro

Teams on IFS Cloud often want Claude for two things: reading long documents such as maintenance manuals, contracts and supplier quality packs, and reasoning over structured data like project costs and work orders. Claude is well suited to both, as long as it gets clean, scoped access to IFS.

The modern way to give it that access is the Model Context Protocol. An MCP server describes a handful of IFS tools, and any MCP-capable Claude client can discover and call them, with your organization choosing what is exposed.

The other path is embedding Claude in your own application through an API. Which path fits depends on who is asking questions, where the data may be processed, and how much control you need.

## Use Cases

### Ask questions across work orders and manuals

**Description:** Combine IFS work order history with uploaded OEM manuals so a technician or planner gets a grounded answer with references to both.

### Contract and supplier document review

**Description:** Claude reads long agreements or certificates of conformance and checks them against the purchase order terms pulled from IFS.

### Project controller analysis

**Description:** Through read-only tools, Claude explains movement in committed and actual cost per project element and drafts the commentary.

### Engineering change impact summary

**Description:** Given a part revision, Claude gathers where-used data and open orders from IFS and writes an impact summary for the change board.

### Integration developer assistant

**Description:** Developers use Claude with the projection metadata to draft OData queries and mapping code, speeding up interface work.

## System Surface

OData v4 projections behind IFS Cloud's REST layer are the main data source for an MCP server. The API Explorer helps you discover what your release exposes.

IFS IAM issues OAuth2 tokens, so the MCP server authenticates as a service client or, better for per-user data, uses delegated tokens so IFS permissions follow the person.

Documents attached in IFS Document Management can be fetched through document-related projections where enabled, then passed to Claude as context. Exact entity names vary by version.

Event actions and Connect can notify an external listener when records change, which lets a Claude-powered process react instead of polling.

## How It Works

### Decide the surface

**Detail:** Pick between an MCP server used from Claude clients, a custom app using the API, or both. MCP suits exploratory users, an app suits a fixed workflow.

### Define read-only tools

**Detail:** Start with item lookup, order status, work order search and project cost summary, each returning compact JSON.

### Implement the server against projections

**Detail:** Write the tool handlers that call IFS with a scoped identity, enforce limits and return clear errors that Claude can act on.

### Pick the model hosting

**Detail:** Use the Anthropic API directly, Claude on AWS Bedrock or on Google Vertex AI depending on your cloud contracts and region needs.

### Add document context carefully

**Detail:** Fetch only the documents needed for the question and cite them in answers so users can verify.

### Review usage and tighten

**Detail:** Audit which tools are used, remove unused ones and add narrow tools where users keep asking the same question.

## Example Flows

### Work order troubleshooting with manuals

**Trigger:** A planner asks why a pump keeps failing.

**Steps:**

- Claude calls the work order search tool for the asset.
- It retrieves the last failures and parts used.
- It reads the relevant manual section supplied as context.
- It answers with the pattern, the likely cause and the source references.

**Result:** A reviewed, cited starting point for the maintenance plan.

### Change impact brief

**Trigger:** An engineer submits a part revision for review.

**Steps:**

- Call where-used and open order tools for the part.
- Summarize affected products and quantities on order.
- Draft the brief for the change board.

**Result:** The board starts from a factual summary instead of a blank page.

## On Prem

Anthropic's hosted models are not available as a fully on-prem, air-gapped deployment. Customers with strict boundaries typically choose one of two designs. Either use Claude through a cloud region contractually acceptable for the data class, with the MCP server running inside your network and sending only the minimal tool results, or keep everything local with an open-weight model such as a Qwen or Llama class model on vLLM. Because the tool layer is MCP, you can keep the same IFS server and swap the model client later.

## Security

- Run the MCP server inside your network and expose only the tools you intend, with parameter validation.
- Prefer delegated, per-user tokens so IFS permission sets still decide what each person can see.
- Return minimal fields in tool results, because everything returned becomes model input.
- Log every tool call with user, arguments and result size.
- Confirm data residency, retention and training terms for your chosen API, Bedrock or Vertex route.

## Options

### MCP server plus Claude clients

**Fit:** Analysts and planners who explore data interactively.

**Effort:** Short for a read-only pilot.

### Custom app on the Anthropic API, Bedrock or Vertex

**Fit:** Fixed workflows with a defined UI.

**Effort:** Moderate, including auth and evaluation.

### Netray-built private deployment

**Fit:** Data that cannot go to any external model service.

**Effort:** Larger, with GPU sizing and model evaluation.

## Pitfalls

- Shared service accounts that ignore the asker's IFS permissions.
- Returning full projection payloads and wasting context on irrelevant fields.
- Feeding whole manuals when a retrieved section would do.
- Assuming a cloud region choice satisfies export control rules without legal review.
- No citations, so users cannot tell grounded answers from guesses.

## Faq

### Does IFS Cloud have a Claude connector?

**Answer:** Check current directories and the IFS ecosystem, since offerings change. If none exists for your needs, build an MCP server over IFS OData projections, which any MCP-capable Claude client can then use with the tools you choose to expose.

### Should we use Bedrock or Vertex instead of the direct API?

**Answer:** Choose based on where your contracts, identity and data controls already live. If your organization is standardized on AWS or Google Cloud, using Claude there can simplify procurement and networking. Behavior is similar, but features can lag, so confirm what you need.

### Can Claude run fully inside our network?

**Answer:** Not as a self-hosted model. If you need strict on-prem operation, use an open-weight model locally and keep your MCP style tool layer. That way the integration work carries over even though the model changes.

### Will Claude respect IFS permissions?

**Answer:** Only if you design for it. Using delegated tokens per user makes IFS enforce permissions. With a shared service account, your tool layer must implement the filtering, which is easy to get wrong.

### Where should we begin?

**Answer:** Build three read-only tools around a question your team asks daily, test them against known answers in Aurena, then add document context. Keep writes out of scope until the read experience is trusted.

## Search terms

ifs cloud claude, claude for ifs erp, claude mcp ifs, connect claude to ifs cloud, anthropic ifs integration, claude bedrock ifs, ifs odata claude, claude erp integration manufacturing, claude on-prem alternative ifs

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