# MCP integration that gives Claude and ChatGPT controlled access to your systems

MCP integration is the work of exposing a company's systems to AI assistants through the Model Context Protocol, so Claude or ChatGPT can answer from live records inside each user's permissions. Goldenroz builds custom MCP servers and installs its own on-prem MCP gateway, with read-only access, approved queries, PII masking and an audit log. Projects start at $25,000, fixed price per phase after a free one-hour scoping call.

**From $25,000, fixed price per phase**

Updated: 2026-11-14. By Shai Goldenberg.

## What usually goes wrong

- **Staff paste records into a chat window:** People copy customer lists and financial exports into an assistant because it has no connection to the source. There is no log of what left the company.
- **A community MCP server with full access:** A developer installed an open-source server and gave it a database user with broad rights. It works, and it lets the model run any query against production.
- **One shared service account:** Every question runs under the same credentials, so a junior employee can ask the assistant for data they could never open in the ERP.
- **On-prem systems the cloud cannot reach:** The ERP or SQL Server sits behind the firewall with no public API. Hosted connectors from the assistant vendor do not cover it.

## What is included

- **Tool design:** MCP development starts with a short list of named tools with typed inputs and clear descriptions. Models choose tools by reading descriptions, so wording is tested against real questions.
- **Custom MCP server or gateway install:** A server written for your internal system or product, or our gateway with its 13 connector types for ERP, CRM, SQL and Microsoft 365.
- **Authentication per user:** Each request carries the identity of the person asking, and the server returns only what that person may already see in the source system.
- **Approved query catalog:** Parameterized queries reviewed by your team. The model selects one and fills in values. It never writes SQL.
- **Masking and audit log:** Personal fields are masked before results reach the model, and every call is recorded with user, query and returned rows.
- **Client setup and handover:** Connection of the server to your Claude or ChatGPT workspace, a Docker Compose deployment inside your network, and documentation for IT.

## Claude integration and ChatGPT integration services

Claude integration through MCP means staff ask a question in the Claude app and the answer comes from your records, with the query visible in the log. Because MCP is a shared standard, the same server supports ChatGPT, so our ChatGPT integration services reuse the work instead of repeating it. Which plan and client features you need depends on your provider agreement, and we confirm that during scoping.

Some teams need the model inside their own software instead of inside an assistant app. That is a direct API project: OpenAI integration or the Claude API called from your backend, with your own interface on top. It shares the same safety rules, and it is covered under [AI integration services](https://goldenroz.com/ai-integration/) and, when the model takes actions, under [AI agent development](https://goldenroz.com/ai-development/).

## MCP consulting before you build

Not every company needs a build on day one. MCP consulting is a short engagement where [Shai](https://goldenroz.com/about/shai/), our CTO, reviews the servers you already run or plan to install, the credentials they hold and the data they can return. The output is a list of risks by severity and a recommended design. The same applies to Claude consulting for teams deciding how to roll out the assistant with company data.

Shai built and owns an on-prem MCP gateway of about 27,000 lines with 13 connector types. It was designed around the failure modes above: read-only access, approved queries only, per-user permissions, PII masking, an audit log and prompt-injection defenses that treat returned data as untrusted text.

## What it costs

Each phase has a fixed price agreed before work starts. The price depends on the number of systems, the number of user roles and whether a custom MCP server is needed.

- AI integration project: From $25,000 (Fixed price per phase)
- Larger multi-system platforms: $100,000 and up (Scoped per project)
- Scoping call: $0 (Ends with a written scope and a price)
- Model subscriptions and usage: Paid to the provider (Your own Claude or ChatGPT accounts)

## Who it is not for

- Giving a model a database password and free-form SQL. We do not build that.
- Write access to production systems on day one. Integrations start read-only.

## Questions buyers ask

### What is MCP integration?

It is connecting business systems to AI assistants through the Model Context Protocol, an open standard for giving a model controlled access to tools and data. After the integration, a user asks a question in Claude or ChatGPT and gets an answer from live records, limited to what that user may see.

### How much does MCP server development cost?

Projects start at $25,000 and are fixed price per phase after a free one-hour scoping call. The price depends on the number of systems, the number of user roles and whether a custom server is needed. Larger multi-system programs run $100,000 and up.

### Does a Claude integration send our database to Anthropic?

No. The server runs on your infrastructure and your databases stay there. Only the results of approved queries, after masking, are sent to the model as context, under your own agreement with the provider.

### Can one MCP server work with both Claude and ChatGPT?

Yes. Both support MCP connections, so one server serves either client. Setup steps and plan requirements differ by provider, and we check them for your accounts during scoping.

### Should we use an open-source MCP server or build a custom one?

An open-source server is fine for a developer's own tools. For company data, check what credentials it holds, whether it allows free-form queries and whether it logs calls. If it fails those checks, a custom server or a gateway with approved queries is the safer route.

---
Page: https://goldenroz.com/mcp-server-development/
Book a free one-hour call: https://calendly.com/daniarozin/daniel-rozin-consultation-1-hour-clone
