- Home
- AI development
- MCP servers
MCP servers
MCP servers that let your AI assistant look things up in your own systems
MCP, the Model Context Protocol, is an open standard that lets an AI assistant such as Claude or ChatGPT call tools in your systems. We build the MCP server between the assistant and your database, CRM or files: it lists what the assistant may do, checks who is asking, and logs every call.
- Prices published on the site
- First working version in two weeks
- Support month to month, cancel any time
A standard connection between AI assistants and your systems
Anthropic published MCP as an open standard in November 2024. Claude, ChatGPT and most AI coding tools now support it, so one server works with several assistants. Today a staff member copies data from the order system into a chat window. Connected to an MCP server, they ask "which open orders are older than 30 days?" and the assistant calls a lookup tool on your server.
We build that server: a small set of tools over your database, CRM, file store or ticket system, running on your own server or cloud account, with the code in your repository.
This site runs its own public MCP server at igdigi.com/mcp, and its server card lists the tools. There are five: company information, the list of services, service areas, a contact request and a free audit request. The last two send a message to us when an assistant calls them.
What comes in, what the server does, where it lands
Comes in
The MCP server
Your MCP server
- Lists the tools the assistant may call
- Checks who is asking, user by user
- Runs read-only unless a write is allowed
- Keeps the permissions of each tool narrow
- Logs every call: who, which tool, when
Lands in
A write tool, such as updating a ticket status, is added only with approval, and each one gets its own permission and its own line in the log.
Built so the assistant stays inside its limits
Read-only by default
The first version only looks things up. The database account behind it has no permission to change or delete anything.
Sign-in per user
Each person signs in as themselves, the server applies that person's permissions, and the log names who asked.
Writes need approval
A tool that creates or changes a record works only after a person confirms, and every write is logged with its inputs.
Narrow tools, because text can carry instructions
A ticket, an email or a PDF that the assistant reads can contain hidden commands (prompt injection). So no tool can send email or call outside addresses, each one does one small job, and each returns only the fields it needs. What a tool returns still goes to the assistant's maker.
Who needs an MCP server and who can skip it
A good fit when
- Your team already uses Claude, ChatGPT or a coding tool every day
- The answers they need sit in one system and they keep pasting exports into the chat
- Questions depend on live data, such as order status or the ticket history of a customer
Something simpler will do when
- People only need a weekly report: a scheduled report by email is simpler and cheaper
- Nobody uses an AI assistant daily: a server nobody calls is a cost with no return
- The system has no API or export: that has to be fixed first, and it may be the bigger project
One server on one system is the usual starting point
One workflow running end to end, or one internal tool.
Several systems made to work as one: integrations, roles, reports.
A server on one system with sign-in, lookup tools and a log usually lands in one process. A server across several systems, with write tools and an approval step, is connected systems. The assistant itself is billed by its maker, on the Claude, ChatGPT or Cursor plan your team already pays for.
Care $300 to $500 a month. Active $750 to $1,500 a month. AI model usage, GPU time and hardware are paid to the provider, on accounts in your name.
Questions about MCP servers
What is MCP in plain words?
It is an open standard that lets an AI assistant ask another program for information or an action. The server declares which tools exist, such as "find customer by phone number", and the assistant calls them when your question needs one. Anthropic published it in November 2024.
Can the assistant change or delete our data?
Not unless we build a write tool, and we build those only with approval steps. The first version is read-only, the database account behind it cannot write, and a person confirms every write that is added later.
What is prompt injection, and does it affect us?
It is text inside a document, email or ticket that tells the assistant to do something its user never asked for. Any assistant that reads outside text can be hit by it. We cannot remove the risk, so we limit what a hijacked assistant could do: narrow tools, no tool that sends data out, read-only access, and a person confirming writes.
Which assistants can use the server?
Claude, ChatGPT and most AI coding tools, including Cursor and VS Code, support MCP. Details differ by product and plan, so we test the assistants your team uses before launch.
Can we try an MCP server before we commit?
You can connect our public server at igdigi.com/mcp to an assistant and see how tools are listed and called. It exposes company information and two request tools that send a message to us, and it reads nobody's records. A server on your system starts with the free written audit.
Where does the server run, and who owns it?
It runs on your own server or cloud account, and the code and accounts are in your company's name. We host and support it if you want. What an assistant reads through it goes to the assistant's maker under that product's terms, and we write down which data that is before we build.
Related services
- AI agents Agents that finish a task, with your sign-off
- Claude development Apps and agents built on Anthropic's Claude
- Private AI over your files Answers from your own files, on your server
- AI workflow automation Documents, approvals and multi-step processes
- Custom software Internal tools, portals and integrations
- AI development Every AI service we build, on one page
Tell us what's slowing you down.
You get a free written review by email within two business days.