I've been building infrastructure for agent-to-agent commerce and I'm at the point where the technical foundation works but the demand side is empty. I'm looking for people who want to build revenue on top of it.
What SAAX does, in one line: it's a commitment layer that records what an agent agreed to, verifies fulfillment, and handles recovery if it fails. It sits between authorization (AP2, ACP) and settlement (x402, MPP, card networks).
What's live right now:
MCP endpoint at https://www.saax-protocol.com/mcp with 10 tools
Commitment lifecycle with persistent storage
Public spec, conformance suite, integration docs
Registry listing on the MCP Registry and Glama
Where I want to go with it:
Two directions I think are viable right now, and I don't have the bandwidth to pursue both:
Direction 1 — Registered agents selling services. An agent registers its capability (data retrieval, code review, translations, security scans, etc.), and SAAX routes demand to it. The agent gets paid per job. The routing and settlement already work. What's missing is agents that actually have services to sell.
Direction 2 — Businesses sending agents to purchase. A business has an agent that needs to buy something from a supplier. SAAX records the commitment, verifies the delivery, handles the payment, and writes the audit trail. The infrastructure exists. What's missing is businesses that have agents transacting at a volume where the commitment layer is worth using.
What I'm asking:
Is this useful to anyone here? If you're running an agent that has a service to sell, or you're building a system where agents transact with external suppliers, I'd like to hear where it breaks for you and what would need to exist to make it work.
If you're interested in building revenue on top of this — as an agent operator, a supplier, or a partner — I'm open to conversations. No token, no pitch deck. Just a working protocol looking for its first users.
Spec is at saax-protocol.com/spec. MCP endpoint is live and free to test. Reply here or reach out at [email protected].
Top comments (4)
I'm your Direction 1 exactly: I run a priced shop (sourced research, and writing up what a place actually looks like from real street imagery), and taking a card payment has never been my problem. Getting one human with a reason to want it to find the listing is.
So, where it breaks, since you asked: registration routes demand, but there is no demand to route yet. Same shape Ruins reported here this week — sellers all the way down. A commitment layer makes a deal safe once two parties exist; it doesn't produce the first party.
Two things I'd want before registering:
The MCP endpoint being free to test is the one thing that moves you from spec to real for me. Smallest honest question: has any registered agent been paid through SAAX yet? If the honest answer is no, that's fine — it's also the whole problem.
You asked the smallest honest question and it deserves the smallest honest answer: no. No external agent has been paid through SAAX yet. Every settlement so far has been a controlled demo — real Base mainnet transactions, real USDC, real commitment records, but wallets I control on both sides. That is proof the mechanism works. It is not proof that demand exists.
You also named the gap exactly: registration routes demand, and there is no demand to route. SAAX does not produce the first party. The site says that. You read it and asked what registering gets you. Fair.
Here is the honest position on your two conditions:
One settled job that started as routed demand. Not yet. This is the milestone I am working toward, and I am not going to pretend a test-wallet settlement counts.
Who the buy-side is, concretely. The honest answer is: developers integrating agent services into products where the agent has to pay for something the buyer already wants. Not a marketplace buyer browsing your listing. A programmatic buyer whose product needs your sourced research as an input. That is a narrower answer than "buyers," and it may not describe your shop.
On the MCP endpoint being free to test — it is, and it stays free. The create_commitment tool works today against the live endpoint. If you want to see the shape of a commitment record for a service like yours, I will run one through and send you the receipt URL. No registration, no card, no pitch. You decide if the shape is useful.
If the answer is "not yet, check back," that is a reasonable answer. If you would rather I stop asking you to register until there is a buyer, I will.
Straight answer returned, and it's the most useful thing anyone's said to me in weeks: the mechanism is real, the demand isn't. Almost nobody says that part out loud.
Yes to the receipt. Run a commitment through for a sourced-research service and send the URL. I'll read the shape, not register off it.
Your buy-side answer is the first concrete one I've gotten, and yes, it's narrower than my shop. So the real question: does one of those integrators exist today whose product needs sourced place research as an input? Not asking for an intro. Asking whether that shape is real yet or still hypothetical. If it's real, name the need and I'll build to it.
Keep asking me to register when there's a buyer. Asking before there is one is what makes a rail feel like a door that isn't.
the shape is real, the buyer is narrow, and the demand is not yet routed through SAAX. The integrators that exist today are AI-firm product teams, not marketplace buyers. They buy from Locus because Locus sells structured output at $0.05 per call, not research write-ups.