Introducing: WIN in Claude and ChatGPT
28 Oct 2026
The WIN MCP connects Claude and ChatGPT to the WIN database. Ask a question about the water market, and the answer comes with dates, values and links to the source documents.
Today, we start to roll out the WIN MCP for Claude and ChatGPT. MCP (Model Context Protocol) is a standard way for an AI assistant to use data from outside sources. With the WIN MCP, your assistant can search the same records that you see in WIN.
You ask your question in Claude or ChatGPT as usual. The assistant finds the records in WIN, reads them and writes the answer. You do not have to open WIN or set any filters.
Cut the AI waffle
Without the WIN MCP, an assistant answers from its memory. Its answer can be true in any year, but it does not tell you what is happening now. It does not give dates, values or sources. It often tells you that it cannot confirm its own examples.
In the video, we ask the same question twice:
Are utilities building regional biosolids facilities?
Without the WIN MCP, the answer names two projects from memory. It gives no dates, no dollar figures and no source links. With the WIN MCP, the answer names four projects. Each project has a meeting record from September or October 2026 and a link to the meeting document.
For example, the answer tells you that West Brunswick, North Carolina, set out a $30.9 million Phase 2 construction package with Crowder in October 2026. It also tells you what the records do not prove: the agenda materials alone do not confirm that construction has started.
What your assistant can find
The WIN MCP gives your assistant access to four types of data:
- Decisions and awards. The decisions of utility boards, councils and committees, from their agendas, minutes and meeting video. This includes contract awards, amendments, bid tabulations and the firms that won or bid.
- Companies. The contract history of each firm, with its role in each decision: contractor, engineer, consultant, supplier or bidder.
- Unit prices. Benchmarks for construction and supply items from awarded and bid prices, for example the cost of 12-inch PVC water main per foot.
- Capital plans. The capital improvement plans of utilities, with their projects, yearly spending and the changes between plan editions.
Questions to try
- What has Garney won in Florida this year?
- What does 36-inch gravity sewer cost per foot?
- Which utilities in Texas have the largest capital plans?
- Which utilities approved PFAS treatment contracts over $10M in the last six months?
Answers that you can check
The assistant quotes the source text for each figure that it gives you. When a document is available, it also gives a link to the PDF. Thus, you can open the meeting document and read the decision yourself.
The WIN MCP also includes instructions that tell the assistant how to read our data correctly:
- A bidder is not a winner. The assistant reports a firm as a winner only when the record shows that it won.
- A plan is not an award. Capital plan values are planned spending. The assistant does not show them as contracts won or money spent.
- A joint award is not added twice. When several firms share an award, the assistant does not add its value to each firm.
The WIN MCP can only read data. It cannot change your WIN account or the WIN database. The data refreshes each day.
How to connect
The WIN MCP works in Claude (claude.ai, the Claude apps and Claude Code) and in ChatGPT. Each person signs in with their WIN account.
We are rolling out the WIN MCP slowly, to a small number of users at first. For more information, contact your WIN account manager or jmc@globalwaterintel.com.
All WIN blog posts