MCP, short for Model Context Protocol, is an open standard that lets AI models and tools connect to other software and work with its tools and data. For estimators, it means an AI tool can read your actual estimating data and answer questions about it, instead of relying on whatever you copy and paste in.
Here’s what MCP is, how it works, what to check before you switch it on, and how it works with Cubit Estimating.
What is MCP?
MCP is a shared set of rules that any software company can follow, so AI models and tools can work with its software. Anthropic released it publicly in November 2024. In December 2025, it was donated to the Agentic AI Foundation, a fund under the Linux Foundation co-founded by Anthropic, Block and OpenAI, with support from companies including Google, Microsoft and AWS. That means no single company controls it.
When people talk about “an MCP”, they usually mean an MCP server: the connection a software company builds so AI models and tools can work with its software, following the MCP standard.
Think of a standard drawing legend. Plans come from hundreds of offices, but because everyone uses common symbols, any estimator can read them. MCP does a similar job for software and AI. Build one connection that follows the standard, and any AI tool that understands the standard can use it.
It also helps to know what MCP isn’t:
- It’s not an AI model. It’s the connection between your AI tool and your software.
- It’s not automatic. Nothing connects until you set it up.
- It’s not a replacement for your judgement. What you do with the answer is still your call.
Why does MCP matter to estimators?
General AI models are good at language, but they haven’t seen your rates, your quantities or the jobs you’ve priced. To get a useful answer, you’ve had to export a report or paste sections of an estimate into a chat.
That works, up to a point. It’s slow, it only covers what you remember to include, and the copy is out of date as soon as the estimate changes.
MCP removes that step. The AI tool reads what it needs directly from the software, so you can:
- ask a question in plain language and get an answer drawn from your own data
- follow up with another question, without starting again
- look across more than one job in the same conversation
At Buildsoft, we think this is where AI earns its place in estimating: not guessing or generalising, but working from your numbers.
How does MCP work?
There are three parts:
- Your AI tool, such as Claude Desktop or Codex. This is where you type your question.
- The MCP server, provided by the software company. It decides what the AI tool can see and do.
- Your data, which stays in the software it belongs to.
When you ask a question, the AI tool works out what information it needs, requests it through the MCP server, and uses what comes back to write its answer. Some MCP servers let AI tools make changes; others only let them read.
What’s the difference between a local and a remote MCP server?
MCP servers run in one of two places:
- A local MCP server runs on your own PC, alongside the software it connects to. It needs an AI tool installed on the same PC that can run it, such as Claude Desktop, Claude Code, Codex or Cursor.
- A remote MCP server runs online and is reached over the internet, usually by AI tools used in a web browser.
Not every AI tool can use both. Browser-based AI apps, such as ChatGPT and Gemini on the web, can’t run a local MCP server, so they can’t connect to software that uses one.
Is it safe to connect AI to my estimating data?
It can be, if you know what you’re switching on. Before you connect an AI tool to your business software, check:
- What the connection can do. Can the AI tool only read your data, or can it change it too?
- Who controls it. Can you switch the connection off when you need to?
- Where your data goes. When an AI tool answers a question, the information it reads is sent to that AI provider to produce the answer, under the provider’s terms.
- What it costs. AI tools measure usage in tokens, and bigger jobs and longer conversations use more.
The Australian Cyber Security Centre (ACSC), with partner agencies in the US, Canada, New Zealand and the UK, published guidance on agentic AI in May 2026 that recommends giving AI agents only the access a task needs. And the Office of the Australian Information Commissioner (OAIC) recommends, as best practice, not entering personal information, particularly sensitive information, into publicly available generative AI tools. Estimates can hold client names and site addresses, so check your company’s AI policy before you start.
Also, AI can get things wrong, so check what comes back before you rely on it.
How does MCP work with Cubit Estimating?
Cubit Estimating by Buildsoft now has its own MCP server: the MCP Connector for Cubit Estimating. It’s the AI Estimating feature that gives AI models and tools read-only access to your Cubit Estimating data, so they can analyse jobs, explore costs, build reports and answer questions about an estimate.
- It’s read-only. AI models and tools can read your estimating data, but they can’t add, edit or delete anything in Cubit Estimating.
- It’s a local MCP server. It works with AI tools that can run a local MCP server on your PC: Anthropic’s Claude, through Claude Desktop and Claude Code; OpenAI’s GPT, through Codex; and Cursor.
- It’s part of AI Estimating, which is available on every edition of Cubit Estimating.
The MCP Connector for Cubit Estimating is read-only, and that’s intentional. AI models and tools can read your data, but they can’t create, modify, or delete anything in Cubit Estimating, so it protects your data integrity. That follows the least-privilege principle in the ACSC’s guidance: AI gets only the access it needs, and every decision stays with you.
For example, ask a question about a job:
“How much have we allowed for plasterboard on level 2 of [job], and how does it compare with [other job]?”
You’ll get the answer in plain language, with the items and figures behind it. Then follow up, such as asking which items drive the difference.
You can also use it to check an estimate before you submit it, check a finishes schedule against your estimate, draft a client quote, compare two versions of a job, and more. You’ll find some examples, with prompts to try, in the Cubit Estimating 2026 October Release.
Frequently asked questions
Is MCP only for Claude?
No. Anthropic created MCP, but it’s now an open standard under the Linux Foundation, and AI tools from many companies support it.
Do I need to be technical to use MCP?
Not usually. Setting it up is typically a few steps: switching on the connection in your software, then connecting your AI tool. For Cubit Estimating, follow the setup guide.
Can an AI tool change my data through MCP?
It depends on the MCP server. Some allow changes and some don’t. The MCP Connector for Cubit Estimating is read-only, so AI tools can’t change your estimates.
What’s the difference between MCP and an API?
An API is built for developers, who write code to connect one system to another. MCP is built for AI tools: it describes what software can do in a way an AI tool can understand and use when you ask it a question.
Does using MCP cost anything?
Your AI tool uses tokens with its provider each time it reads data and writes an answer. With the MCP Connector for Cubit Estimating, each lookup uses tokens through both Cubit Estimating and your linked AI tool.
See what MCP can do with your estimates
The MCP Connector for Cubit Estimating is available now with the AI Estimating module, on every edition of Cubit Estimating. Book an AI Estimating demo, and we’ll show you how it works on the kind of jobs you price.
