Almost everything a buying agent needs, except the one thing that would tell it whether to buy.

The list since spring is long. Google's Universal Commerce Protocol added Catalog and Cart in March, so an agent can pull variants, inventory and pricing from a merchant in real time. Google donated the Agent Payments Protocol to the FIDO Alliance on 28 April. The Linux Foundation stood up the x402 Foundation in April with Amazon, Google, Stripe, Visa and Mastercard as premier members. Mastercard launched Agent Pay for Machines on 10 June, agent-to-agent payments with more than thirty partners, and Visa announced an Agentic Registry the same day. The Model Context Protocol shipped its largest specification release on 28 July.

Read the specifications side by side and a pattern shows up. Every one of them answers four questions: who is this agent acting for, what is for sale, what does it cost, and how does it pay. Identity, catalog, price, settlement.

Not one of them has a field for the economic value the product delivers to the buyer. The buying agent can now read every price on the internet. It cannot read what any of those things are worth.

Procurement software. Quickly, and with price data.

Vertice bought Vendr on 1 June to assemble a dataset of 250,000 contracts across 32,000 vendors, explicitly to train an autonomous negotiation agent. Coupa's August release shipped an MCP server with more than thirty tools, alongside a benchmarking service that scores a quoted price against market indicators. Zip has had a price negotiation agent since last October that coaches buyers through software renewals using their purchase history.

This heralds the arrival of the buying agent. It will turn up to your renewal with a benchmark for what everyone else paid. The open question is what the seller's side turns up with.

It runs the only analysis it can. Cost of ownership. And thereby misses value. The full equation has a denominator, the cost or price, and the numerator, a value.

G2 surveyed 1,038 B2B software buyers in June of 2026. Sixty-one percent are using or planning to use AI agents in the buying process. The most common job they give the agent is comparing total cost of ownership, at 51%. In the same survey, 49% said finance had reversed a purchase the buying team had already approved in the past year.

Last week I wrote about the mismatch between the question the seller answers, which is benefit, and the question the buyer runs, which is cost. Agents make that mismatch structural. When the only machine-readable input is the price, the comparison is a price comparison, and the vendor with the lowest number wins by default. A value model that lives in a spreadsheet, on a slide, or behind a login in a vendor's platform is not an input. The agent never sees it.

The Value Project launched in May with five open JSON schemas. ValueModel and CustomerVariables describe what a product is worth to a specific customer and why. PricingModel, DealConfiguration and InvoiceStatement describe what that customer would owe and why. The schemas are published under Apache 2.0. The specifications are published under CC BY 4.0. Anyone can implement them without asking us.

Since launch, the value-model specification has been revised twice, adding sourced benchmark claims and customer criteria written for the buyer's audience rather than the seller's. The pricing-model specification is on its second major version. Both repositories are public on GitHub.

valueIQ published its own value model and its own pricing model in the format: 30 variables, 10 value drivers, 9 benchmarked improvement claims, 4 tiers. An agent can retrieve it, put in its own numbers, and compute the result. In June I used the format to compare three digital asset management vendors, Bynder, Frontify and Papirfly. The exercise showed something a feature comparison never does. Each vendor creates value from a different reference point, and once the models are public that difference is legible to anyone, including (especially) a machine.

They shipped connectors.

valueIQ has a well-defined and complete set of MCP servers. On 8 September, Cuvama announced Cuvama MCP, which lets Claude, ChatGPT, Copilot and Gemini read and update a value plan inside Cuvama's platform. Minoa has offered a remote MCP server since the spring that serves its customers' value library to agents. All these are sensible product moves, and I expect the rest of the category to follow.

They are also a different thing from a standard format, or schema, that anyone can use. A connector lets an assistant read a model that lives inside one vendor's system, for that vendor's customers, in that vendor's structure. A format or schema lets any system read any model, regardless of who built it. The buying agent at your prospect does not have a login to your value selling platform. It needs the model published somewhere it can fetch, in a structure it already understands.

As far as I can find, no value selling vendor other than valueIQ has offered one. Nine companies sell value modelling as a product. Only valueIQ has made the value model legible and portable.

Implement the schemas. Publish your customers' value models, or start with your own.

This is a genuine invitation to Minoa, Cuvama, Ecosystems, Symbe, ValueCore and Mediafly, and to anyone else building in this space. The schemas are free. You do not need valueIQ to use them. Register your implementation in the public IMPLEMENTATIONS file in either repository. At ten independent production implementations, governance moves from valueIQ to an independent technical steering committee, and the roadmap after that is the Linux Foundation, where the payment protocols already live. We wrote that transfer into the governance document because a standard that only its sponsor uses is not a standard.

The alternative is that the buying agents arrive, find catalogs and prices and nothing else, and every vendor in every category gets compared on cost by a machine. That outcome is bad for every company that sells on value.

Does an MCP connector make a value model machine-readable?

Inside the vendor's platform, for that vendor's own customers, yes. To a buyer's agent with no account there, no. Machine-readable in the sense that matters means retrievable and computable by a system that did not create the model.

Do I need valueIQ to use The Value Project schemas?

No. The JSON schemas are Apache 2.0 and the specifications are CC BY 4.0. Any developer, vendor or platform can implement them. valueIQ's products generate and consume conformant models natively, and that is the extent of the relationship.

What is a value page?

The companion to a pricing page. A pricing page tells a reader what a product costs, in a form any human or agent can parse. A value page tells them what it is worth, in the same kind of form. The value-page specification in the value-models repository defines what one looks like.

What happens if a buying agent has my price and no value model?

It compares you on the input it has. If the only structured data about your product is the price, the comparison is a price comparison, and the buyer's finance team gets a cost-of-ownership table with no benefit column. Or in some cases it will hallucinate your value. Neither is good.

The protocols will not add a value schema for you. The procurement agents will not ask for one. The only party with an interest in a buying agent understanding what your product is worth is you.

We published ours. See valueIQ's own value model, in the open format, at valueiq.ai/value-model.