Connect · Example build

AI connectors (MCP): give ChatGPT, Claude and Copilot secure access to your systems

Gives ChatGPT, Claude or Copilot (AI tools you already pay for) secure access to your ERP, database and files, behind your single sign-on.

Your team already pays for ChatGPT, Claude or Microsoft Copilot, but those tools can't see your ERP, your database or your files, so they can't do real work. AI connectors fix that. We build Model Context Protocol (MCP) servers for your systems, so the AI tools your team already uses can look things up and take approved actions in them.

Every connector signs people in through your existing single sign-on, takes roles from your directory groups, and logs every request under a named user. Remove someone from SSO and their access ends everywhere.

The problem

You pay for ChatGPT, Claude or Copilot, but they can't see your ERP, your database or your files, so they can't do real work.

Signs you need it

  • People copy and paste data from your systems into ChatGPT or Claude to get work done.
  • You want AI tools to work with company data but security won't approve shared API keys.
  • Different roles need different access: sales shouldn't see payroll, interns shouldn't post entries.
  • You need an audit trail of what AI tools accessed and on whose behalf.

How it works

  1. 1

    We build an MCP server for each system: your ERP, a database, a file store or an internal tool.

  2. 2

    People sign in with your existing single sign-on (Microsoft Entra ID, Okta or Google Workspace). No new passwords, no shared keys.

  3. 3

    Roles come from your identity provider's groups, so each person can only reach the data and actions their role allows.

  4. 4

    Every request is checked against the person's role and logged under their name. Remove someone from SSO and their access ends everywhere.

Try the demo

Runs on sample data in your browser. Nothing you do here is sent anywhere.

AI CONNECTORS · MCPSSO · MICROSOFT ENTRA ID

Ask Claude about stock. It reaches your ERP through a connector, within the permissions on the right.

SSOSigned in as j.ortiz@example.com

Microsoft Entra ID · group SG-Purchasing

Read stock. Draft POs; a manager approves.

Connectors · Claude

  • ERP

    Read: Inventory, orders, vendors

    Create purchase orders: Needs approval

  • Sales database

    Read: Invoices, customers

  • SharePoint

    Read: Policies, price sheets

Audit log

Every call is logged here.

Simulated with sample data and fictional users. Works the same way in ChatGPT, Claude or Copilot.

Example uses

ERP and accounting

Look up customers, orders, invoices and inventory from inside Claude or ChatGPT, with write actions such as drafting a purchase order allowed only for roles you choose.

Databases and reporting

Read-only access to a documented set of tables, so analysts can ask questions in their AI tool instead of filing report requests.

Files and documents

Search and read SharePoint or file-share documents through the AI tool, respecting the permissions people already have.

Internal tools

Wrap an internal app or API as an MCP server, so routine actions can be done from the AI tool with approval where it matters.

How we build it

We build one MCP server per system, each exposing a small set of well-described tools (search customers, get order, draft invoice) rather than raw database access. Narrow tools are easier to secure, test and audit.

Sign-in uses OAuth through your identity provider (Microsoft Entra ID, Okta or Google Workspace), following the MCP authorization specification. No new passwords and no shared keys. Roles map from your directory groups, and every tool call is checked against the caller's role before it runs.

Read-only is the default. Write actions are opt-in per role, can require approval, and are logged with the user, the inputs and the result. We then set the connectors up in the AI tools your team uses.

Where people stay in charge

Single sign-on and role-based access from your identity provider. Read-only unless a role allows a write action, approvals where you want them, and every call logged to a named user.

How it’s tested

Every build ships with an eval suite built from your real examples, re-run on every change, before and after launch.

Every role gets permission tests: each one tries the things it shouldn't be able to do, and the test fails if it succeeds. Task tests cover the tools your team uses most.

Both suites run on every change, so a new tool or an upgraded dependency can't open a gap in access control.

What you get

  • MCP servers for your chosen systems
  • SSO through your identity provider (OAuth, per the MCP authorization spec)
  • Role-based permissions mapped to your directory groups
  • A per-user audit log
  • Setup in the AI tools your team uses
  • The source code and documentation, deployed in your environment. You own both.

We start with paid discovery: we learn how the work is done today, look at your real documents and data, and agree on what to build and what it has to get right. Then we design and build it, as a project with a not-to-exceed cap or on a Lab retainer. It runs on leading commercial models under terms that don't allow training on your data, or on a private model in your own environment.

Questions

What is an MCP server?

MCP (Model Context Protocol) is an open standard that lets AI tools such as Claude, ChatGPT and Copilot connect to outside systems. An MCP server exposes a set of tools, like 'look up an order', that the AI can call on a person's behalf.

Can ChatGPT, Claude or Copilot connect to our ERP?

Yes, through an MCP server built for your ERP. People sign in with your single sign-on, and each person can only reach the data and actions their role allows.

How do single sign-on and roles work with MCP?

The connector uses OAuth through your identity provider, per the MCP authorization specification. Roles come from your directory groups, every call is checked against the caller's role, and access ends when someone is removed from SSO.

Can the AI change data, or only read it?

Read-only by default. Write actions are enabled per role, can require a person's approval, and every one is logged to a named user.

Last updated October 20, 2026

Need something like this?

Tell us what you’re working with. We’ll reply with a few questions and a time to talk.

Start a build like this

See all builds →