29 May 2026
How to generate quotes with an AI agent — and why the model must not price them
An increasingly common wish: “I want to describe a job in Claude Desktop and get a quote.” It's doable — but only if a tool computes the price, not the language model. Here's how.
We keep hearing the same wish from small make-to-order businesses — joinery, metalwork, installations: “I want to describe a job in Claude Desktop (or ChatGPT) and get a quote.” Picture a joinery that builds custom windows and doors: the owner would like to dictate “three wooden windows 120×140, double glazing, one fire door” and get a finished quote.
It's doable. But there's a trap that separates a useful solution from an expensive mistake.
The trap: the model must not price things
A language model is great at confidently guessing. Ask it to “estimate” the price of a window and it will give you a tidy number — often a wrong one. In quoting, a wrong price isn't a typo; it's a lost deal or a loss on the deal.
The rule is simple: the AI gathers parameters and writes the text, but a tool computes the price — deterministically, from your price list. Same result every time, verifiable.
The solution: an MCP connector for quotes
Between Claude Desktop and your price list we put an MCP connector — a small server with a few tools:
calculate_quote— input is the line items (type, dimensions, material, glazing, quantity, options); output is priced line items and a total (with and without VAT). Your rules do the math, not the model.create_quote— saves the quote, assigns a number and returns a PDF with your logo, terms and validity.- optionally
customers— a simple directory for repeat clients.
The flow: you describe the job in Claude Desktop → the agent calls calculate_quote, then create_quote → it returns a PDF and a number. No new app to build and learn.
The hardest part is your price list
Honestly: the AI layer is the easy part. The value and most of the work are in digitising the pricing logic. If your price list is in a spreadsheet, great. If it's “in the master's head,” the first step is to capture it into rules — and that's the biggest benefit of the project, AI aside.
A draft, not blind automation
For make-to-order work the agent prepares a draft quote that a human approves. Measurements and feasibility stay with the craftsperson; AI removes the typing, the math and the formatting, not the judgement.
Customer data is processed in the EU (GDPR), behind a secure token.
The result
A quote produced in a minute instead of half an hour, consistent, numbered and stored. It's a concrete example of our AI integration service: we wire AI into your existing process rather than sell you another tool.