If you read this and the only thing you take away is "another acronym to ignore," I get it. But MCP — short for Model Context Protocol — is quietly becoming the most consequential infrastructure shift in construction finance tech this year, and it's worth five minutes.
The problem MCP solves
If you've ever asked your team "which vendors did we spend the most with on Job 4421?" — and then watched the next 40 minutes burn on exporting from your ERP, pivoting in Excel, and pasting into a deck — you've experienced the problem. Your data is in your Accounting system, your CRM, your email, your project-management tool, your email. AI assistants can't actually read any of it. So the AI side of "AI-powered finance" mostly produces text-shaped summaries of data you already had to assemble manually.
MCP changes that. It's an open standard, introduced by Anthropic in late 2024, that lets software platforms expose their data and workflows in a way that AI assistants can consume on your behalf. Think of it as the cable that finally lets Claude (or ChatGPT, or whatever you use) actually look inside your AP system, your project files, your accounting system, you PM system, your CRM, your email, your calendar — and answer questions about them in plain English.
Why this matters for construction
The construction software stack has historically been stitched together by humans. Every reconciliation between the Accounting and the PM system, every cross-check between job costs and the project schedule, every "did this sub send the invoice that matches the pay app?" — those have been manual workflows because the systems don't talk to each other natively.
MCP doesn't eliminate that fragmentation. But it does mean an AI assistant sitting on top of all those systems can do the cross-system work — pulling, joining, summarizing, flagging — without you ever opening Excel.
It's already happening in construction tech
In the last 12 months, JobTread, Procore, and Autodesk have all shipped MCP servers. AI assistants can now directly query project management, RFIs, submittals, and design files in those systems. The back-office side — AP, accounting, document automation — is starting to catch up too. (At Beiing Human, we shipped ours last week. More on the specifics at the end if you're curious.)
A concrete example of what changes
Pre-MCP, a controller's question to "the AI" was usually a polite way of asking it to summarize something you'd already done. Post-MCP, it's a question you ask once and get an answer to.
Examples of questions you can now ask an AI that's connected to your AP system via MCP:
- "Pull every invoice from our top equipment rental vendor last quarter and break it down by job."
- "Show me all unapproved invoices over $5,000 aging more than 30 days."
- "Which vendors have we overspent on, relative to budget, on Job 4421?"
The AI doesn't summarize a CSV you exported. It calls your AP system, runs the query, joins it against the relevant job data, and gives you the answer. In one conversation, across multiple connected systems.
What to do about it
- Ask your software vendors if they have an MCP server, or have one on their roadmap. If they don't, you'll be doing manual reconciliations longer than the people whose vendors do.
- Start with Claude Desktop or claude.ai. It's the most mature MCP client right now. Try connecting one of the MCP-supported systems you already use (Gmail and Slack both have one) and ask it real questions.
- Think about what your job looks like when the cross-system work is free. Most controllers I talk to have built their week around the manual reconciliation. If that goes away, what does the role become? My guess: a lot more strategic, a lot less spreadsheet.
I shared more specifics on what we built and how we think about it in a longer post on the Beiing Human blog: https://beiinghuman.com/ai-accounts-payable-construction-mcp-server/
Happy to discuss in the comments if anyone has questions, or wants to compare notes on how you're approaching AI in construction finance.