Skip to content

All servicesAI connectors

Let Claude, ChatGPT or Codex work in your systems, with a custom MCP server.

Your team already pays for an AI assistant, but it cannot see your data or do anything in your tools. We build a custom MCP server, a connector, that lets it read and act in your own systems. Each person signs in, roles set who can do what, and anything unusual, or above a limit you set, waits for approval.

Example

The problem

Your AI assistant knows everything except your business.

Claude, ChatGPT and Codex are good at reading, writing and working things out. On their own, they cannot see your customer records, your orders or your documents. They cannot do anything in the tools you run on.

So people copy and paste. They pull figures out of one system, paste them into the chat, then paste the answer somewhere else. The answers stay generic, and sensitive data ends up in chats with no control over who sees it.

  • Copying and pasting between the AI chat and your tools
  • Generic answers that do not know your business
  • Sensitive data pasted into chats
  • No control over who can see or change what

Before and after

Copy, paste, repeat, or ask once.

Someone needs to update a customer's order. On the first lane they shuttle data between the chat and the system. On the second, they ask once and approve the change.

Today

5 steps, every time

  1. Open the systemFind the customer and the orderthen waits minutes
  2. Copy the detailsInto the AI chatthen waits minutes
  3. Ask the AIIt only knows what was pasted
  4. Paste the answer backInto the system or an emailthen waits minutes
  5. Hope it was rightNo record of what was shared

Agitech comes in

Connected

One ask, one approval

You ask onceIn the chat you already use

The connector does the legwork

  • Finds the customer
  • Reads the order
  • Drafts the update
in the chat
You approveBefore anything changes
Updated and recordedWho asked, who approved, when
  • No copying and pasting
  • Answers from your own data
  • Every action on record

Example. The steps and waits on the first lane are typical, not measured.

What we build

A custom MCP server, built around your systems and your rules.

MCP, the Model Context Protocol, is an open standard for connecting AI assistants to tools and data. An MCP server is the piece between the assistant and your systems. It decides what the assistant can read, what it can do and who it is acting for.

  • Read

    Answers from your own data

    The assistant can look up the systems you choose: customer records, orders, stock, documents.

  • Act

    Actions in your systems

    Each action is a tool, such as create a quote, update a record or raise a ticket. We build only the tools you ask for, each with clear limits.

  • Sign-in

    A sign-in for each person

    Everyone signs in with their own account, so the assistant only sees and does what that person is allowed to.

  • Control

    Roles and approvals you set

    Roles decide who can read and who can change. Anything unusual, or above a limit you set, waits for a person to approve it.

  • Record

    A record of every action

    Who asked, what the assistant did, who approved it and when, so you can always see what happened.

  • Ownership

    Yours, wherever it runs

    Where it runs is agreed when we scope it. You own the code and the data, with no licence fees.

How it works

From copy and paste to one question in 4 steps.

  1. Tell us the jobs

    Which systems, what the assistant should read and what it should be able to do. You get a fixed price and a date within 48 hours.

  2. We set the tools and rules

    Each tool, who can use it and what needs approval. You do not need technical staff to answer these questions.

  3. We build and test it

    We build the connector and test each tool against your systems before your team relies on it.

  4. Your team connects and asks

    People add the connector to the AI assistant they already use, sign in and start asking. Most builds take 4 to 6 weeks.

Proof

We build connectors, and we run one of our own.

Studio is a connector (an MCP server) that Agitech built and runs. It lets a team run its website from Claude or Codex. Each person has a role, every change gets a preview, nothing goes live without approval and every action is recorded. The code stays in the client's GitHub.

See how Studio works
6
Jobs run from one chat, from content to security
2
AI assistants it runs from: Claude and Codex
1
Approval before anything goes live

Questions

Questions about Claude connectors and MCP servers.

What is an MCP server?

An MCP server is a connector that lets an AI assistant use a company's tools and data. MCP, the Model Context Protocol, is an open standard, so one connector can work with any assistant that supports it.

Why build a custom MCP server instead of using a ready-made one?

Ready-made connectors cover popular apps in a general way. A custom MCP server is built for your own systems, combines them where needed and applies your roles, approvals and limits.

How do I connect Claude to my company's data?

Through a connector. Claude supports MCP, so a connector built for your systems is added like any other Claude connector. Each person signs in, and Claude reads and acts within their role.

Can you build Claude connectors?

Yes. A Claude connector is an MCP server that Claude can sign in to, which is how Claude MCP support works. We build it once, and it also works with ChatGPT and Codex where they support MCP.

Does it work with ChatGPT and Codex?

Yes, where they support MCP connectors. ChatGPT connectors and Codex MCP support depend on your plan and settings, so we confirm what works for your team before we start.

Is it safe to connect AI to our business systems?

It can be, if the connector sets the rules. The assistant can only use the tools we build, and each person signs in with their own account. Roles limit what they can do, and every action is recorded.

Can the AI change our data without asking?

Only within the limits you set. We design each connector so that anything unusual, or above a limit you set, waits for a person to approve it.

Who owns the MCP server you build?

You do. You own 100% of the code, the data and the systems, with no licence fees. You can leave at any time.

What do your AI integration services include?

Everything from scoping to a working connector: the tools, the sign-in, the roles and approvals, the record of actions and the set-up in your team's AI assistant.

How much does a custom MCP server cost?

You get a fixed price within 48 hours of telling us what you need, with no hourly billing. Most builds take 4 to 6 weeks. Connectors that join several systems can take up to 16 weeks.

Show us the copy and paste your team does every day.

  1. You tell us

    Describe the work you want off your team's plate.

  2. We reply in 48 hours

    You get what AI can take over, a fixed price and a delivery date.

  3. You decide

    Go ahead, change the scope or walk away. There is no obligation.