Many have MCPs. Few have systems worth connecting to.

MCP won't be a competitive advantage for long. It's a protocol, and protocols become checkboxes. Just like APIs, SSO, and webhooks all got the same launch hype and then ended up buried in an unread security questionnaire. MCP is like a door. Installing one fast says nothing about what's on the other side of it. Vendors are already racing to announce their MCPs the same way, because shipping the announcement before the product costs nothing today. But that only defers the cost. When the door opens and there’s nothing solid behind it, what gets taxed later is trust. 

‍

Clean data and genuine adoption give models something worth reasoning over. A decade of half-filled purchase requests and shadow spreadsheets gives it the same mess, now reachable by API. Ask it to flag your riskiest supplier contracts and it'll hand you one for a vendor that went bankrupt two years ago, cited with total confidence and formatted with legitimacy. Wrong has never looked this professional. And it's not the model malfunctioning. That's the model working exactly as built, on data nobody maintained. Tack an MCP onto that and you haven't added intelligence. You've added artificial confidence, and confidence is the most expensive thing to get wrong.

‍

Claude, ChatGPT, and Gemini already do things that would've sounded farfetched a few years ago. None of them know your suppliers, your policies, or how your team operates, and no MCP launch is going to admit that in the press release.

‍

‍

The market is optimizing for the wrong milestone

Every procurement vendor's roadmap this year has the same line item: ship the connector, announce the protocol, get to the demo stage. That's a race to say yes when someone asks if you support MCP. It says nothing about what happens the first time someone actually asks it a real question.

‍

The vendors moving fastest here are often the ones with the least reason to want you looking behind that door. A new front door is easy to install. Rebuilding what's behind it is much harder, so most skip it. They add AI to a structure assembled over a decade of acquisitions, workarounds, and features nobody had time to reconcile, then market the door as the achievement.

‍

We made a different bet at Pivot: before building anything an LLM could talk to, we rebuilt the structure itself, the system of record underneath everything else. A door only matters if you already trust what's behind it.

‍

What actually happens when the door opens

Ask it about a renewal and it hands back a supplier recorded under three different names. Ask about total spend and it gives you a confident number that's wrong, because half the invoices never made it into the system in the first place.

‍

In both cases, the model reasoned correctly over what it could see. The problem was what it could see: a foundation nobody finished building.

‍

This is also the part that never shows up in a vendor demo. Every demo runs on the cleanest data in the building, curated, deduplicated, current as of that morning. The real environment, the one your team lives in every day, is rarely that. Once the door is open, the tour doesn't ask permission before showing you the rest of the house.

‍

MCP depends on these two things

Start with the frame, because nothing else matters if this part's rotten: a system of record, supplier data, contracts, invoices, policy, held together in something that can actually bear weight. Not a shared drive. Not a folder somebody insists is "basically organized."

‍

None of that holds up on its own, though. A frame nobody maintains is just a frame slowly coming apart. That's the system of engagement, whether the people living in this house day to day are the ones keeping it current, not whether a schema exists somewhere with their name on it.

‍

Only once both of those are solid does it make sense to let AI actually touch anything: closing a renewal, correcting a record, reclassifying spend without a human hovering over every keystroke. That's agentic action, and it's the one everyone wants to skip straight to, mostly because it's the one that looks best in a demo.

‍

Why the real moat isn’t MCP

Having an MCP will stop being a differentiator the moment everyone has one, which is roughly a press cycle away. Soon "does your platform support MCP" will get the same shrug as "do you have an API," a question people keep asking out of habit, long after the answer stopped meaning anything.

‍

The differentiator was always going to sit behind the door, decided years before anyone wired up a connector: whether the system of record was built to hold structure from day one, or stitched together across a decade of acquisitions and features nobody bothered to reconcile.

‍

That distinction is not a sprint. A vendor can ship an MCP over a weekend, and plenty are already bragging about exactly that in a press release. Untangling a decade of duplicate suppliers, orphaned contracts, and approval chains nobody re-drew after the last reorg takes considerably longer, assuming anyone ever gets around to it.

‍

MCP is a multiplier, not a renovation crew. It does not show up, assess the damage, and quietly fix the plumbing. What's broken behind the door stays exactly as broken as it was. The only thing that changes is how fast, and how publicly, everyone finds out.

‍

That's the bet we made building Pivot: the record first, the door second. Plenty of vendors do it backwards, because backwards demos better. It's also the question every vendor selling an MCP should have to answer in public, under a direct question, not buried in a keynote slide about the future of agentic procurement.