# How to connect Claude to ECI JobBOSS2

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

## Short answer

Connect Claude to JobBOSS2 by running a Model Context Protocol server next to the SQL Server database that exposes a few read-only tools, then adding it as a connector. For data that cannot leave your network, use an open-weight model instead of Claude.

## System

ECI JobBOSS2

## Capability

claude

## Intro

Claude is good at long-document reasoning and careful, structured work, which fits job shop tasks such as reading a 30-page customer spec against your routing, or summarising a quarter of job cost variances. What it lacks, out of the box, is any knowledge of what is in your JobBOSS2 database.

The standard bridge is MCP, the Model Context Protocol. An MCP server describes a handful of tools, and a Claude client (Claude Desktop, Claude Code, or an app you build on the API) can call them. Your JobBOSS2 data stays where it is until a specific tool call returns specific rows.

This page covers how to set that up, the deployment choices for Claude itself, and where the line falls for regulated work.

## Use Cases

### Spec versus routing review

**Description:** Provide a customer PO and drawing notes and have Claude compare them to the quoted routing and material in JobBOSS2, flagging mismatches before the job releases.

### Quarterly variance narrative

**Description:** Claude pulls closed-job cost data through a tool and writes a plain-language explanation of where actual hours diverged from estimates.

### Order desk questions

**Description:** Staff ask about order status, ship dates or inventory through Claude Desktop instead of navigating screens.

### Quote preparation assistant

**Description:** Claude finds comparable past jobs through a search tool and drafts a quote worksheet with assumptions listed.

### Customer corrective action draft

**Description:** From a returned job and its operations history, Claude drafts the factual timeline for an 8D or corrective action response for a quality manager to complete.

## System Surface

- Microsoft SQL Server database: create read-only views and a dedicated login. Typical underlying tables include Job, Job_Operation, Material_Req, Delivery, PO_Header and Quote, with names varying by version.
- JobBOSS2 API, if included in your edition, can back the same tools. Confirm available resources and auth with ECI.
- An MCP server can be written with the official SDKs (TypeScript or Python) and exposed over stdio for a single desktop or over HTTP for a team.
- Report exports (Excel, CSV, PDF) remain useful as a fallback input for ad hoc analysis in Claude without any connector.

## How It Works

### Decide the Claude deployment

**Detail:** Options include Claude apps with a team or enterprise plan, the Anthropic API, or Claude through AWS Bedrock or Google Vertex AI if your IT already standardises on those clouds.

### Build the MCP server

**Detail:** Define tools like get_job, list_open_jobs, get_material_status and search_quotes. Each should have a typed schema and a hard row cap.

### Connect it to SQL views

**Detail:** Use a read-only login scoped to views. Parameterise every query, never concatenate model-provided text into SQL.

### Register as a connector

**Detail:** For desktop use, add the server in the client configuration. For teams, host it over HTTPS with authentication and register it as a remote connector.

### Write tool descriptions carefully

**Detail:** The descriptions are what Claude reads to decide when to call each tool. Say what the tool returns, units, and what it cannot do.

### Test with real questions

**Detail:** Run 30 questions your staff actually ask. Check answers against the application screens before widening access.

## Example Flows

### Pre-release spec check

**Trigger:** A planner is about to release a new job for a repeat customer.

**Steps:**

- The planner attaches the new PO text to Claude.
- Claude calls get_job and get_routing for the quoted job.
- It compares material grade, finish and quantity to the PO.
- It lists differences and asks the planner to confirm each.

**Result:** Catches spec drift before material is cut.

### Month-end margin story

**Trigger:** The controller asks why margin dropped in the month.

**Steps:**

- Claude calls a closed-jobs cost tool for the month and the prior month.
- It groups variances by customer and operation.
- It writes a one-page summary with the supporting job numbers.

**Result:** A draft narrative the controller edits, with traceable numbers.

## On Prem

Claude is a hosted model, so a fully air-gapped JobBOSS2 shop cannot run it locally. The practical split is this: keep the MCP server and database on your network, and decide per data class what may reach a hosted model. Where ITAR, CUI or strict customer NDAs apply, the same MCP-style tool layer can sit in front of an open-weight model served on your own GPUs, so the architecture is reusable. Shops often use hosted Claude for non-sensitive work and the private model for job and drawing data.

## Security

- Review the provider's data handling terms for your plan, including retention and training exclusions, before connecting any customer data.
- Read-only database login, scoped to views, with parameterised queries only.
- Authenticate HTTP MCP servers and map each user to a JobBOSS2-aligned role so people only see what they should.
- Cap rows returned and log every tool call with user and parameters.
- Treat text inside job notes and customer attachments as untrusted, since it can contain instructions aimed at the model.

## Options

### Exports into Claude

**Fit:** Quick wins for analysis and writing with no integration work.

**Effort:** Hours

### Build your own MCP server

**Fit:** Good for a shop with a developer. Straightforward for read tools, careful work for access control.

**Effort:** 2 to 6 weeks

### Netray or another partner

**Fit:** When you want hosted and on-prem model options behind the same tool layer, plus someone to evaluate answer quality.

**Effort:** A few weeks for a pilot

## Pitfalls

- Giving the MCP server a broad SQL login 'just to test' and never tightening it.
- Vague tool descriptions, leading Claude to pick the wrong tool or invent parameters.
- Letting job note fields flow into the model unfiltered, which opens prompt injection from attachments or vendor text.
- Assuming all users should see all customers. JobBOSS2 roles do not automatically carry over to your MCP server.
- Skipping a legal check on cloud processing of controlled technical data.

## Faq

### Does ECI provide an official Claude connector for JobBOSS2?

**Answer:** We are not aware of one that you can rely on, and offerings change, so ask ECI or your reseller. A custom MCP server over read-only SQL views is a practical alternative that works on most on-prem installs.

### Which Claude deployment is best for a manufacturer?

**Answer:** It depends on your existing cloud. Teams already on AWS may prefer Bedrock, Google shops may prefer Vertex, and others use the Anthropic API or Claude plans directly. The MCP server you build is largely the same across them.

### Can Claude run on-prem against JobBOSS2?

**Answer:** No, Claude is accessed as a hosted service. For on-prem needs, reuse the same tool layer with an open-weight model running on your own hardware.

### Is MCP better than a custom API for this?

**Answer:** MCP is a standard way for different AI clients to discover and call tools, so you build once and can use Claude, other assistants and agents. A plain REST API works too but needs a separate adapter per client.

### Can Claude change data in JobBOSS2?

**Answer:** Not unless you build write tools, and we recommend against it at first. Start with read tools and drafts, then add a narrow, approved write path later if the value is clear.

## Search terms

jobboss claude, claude for jobboss2, connect claude to jobboss, claude mcp jobboss, anthropic claude job shop erp, jobboss2 claude desktop, claude bedrock manufacturing erp, claude eci jobboss integration

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