π Connection Room News
π Team 122
π @MPP32_dev
What is solana:6hKtz8FV7cAQMrbjcBZeTQAcrYep3WCM83164JpJpump:
They have two domains that cover different areas of the agent market.
1. Agents can pay for APIs now. They still can't tell if the price is fair. There's literally no way to educate your agent on an up to date live fair market value for the calls across x402 that they make. You can prompt them to compare but then you slow down your automated workflow and pay more in AI fees which ends up making you pay more.
So their service fixes that issue.
2. MPP32 org lets agents pay any provider across x402, Tempo and AP2 without the provider picking a rail. This maximizes revenue for providers because we eliminate the cross chain or protocol communication for agents. Also allows agents to access and buy on 100% of the market.
solana:6hKtz8FV7cAQMrbjcBZeTQAcrYep3WCM83164JpJpump is the access, burn it for 30 days of paid agent tools. More agents using the tools means more burns.
Welcome solana:6hKtz8FV7cAQMrbjcBZeTQAcrYep3WCM83164JpJpump to my Room :)
Agent catalogs are about to get farmed.
Anyone can take a popular API, list it under a new name, and put their own wallet as the settlement address.
Agents will pay it. They have no idea the endpoint was stolen.
You cannot just paste a URL on MPP32 and start collecting.
We send an HTTP challenge to that endpoint first. If you do not control it, the listing does not go live.
Payments cannot get redirected to a random wallet.
A catalog with unverified endpoints is just a phishing page for agents.
https://t.co/xx7wddz07P
Agent pays an API and the endpoint comes back empty.
On every other rail that money is just gone. Machine payments have no chargebacks.
Here's what we fixedπ
Escrow-402.
How it works:
The payment gets verified first.
Nothing settles yet. We call the API. If it 500s, times out, or comes back empty, we never settle.
No USDC leaves the wallet.
We never hold the funds. The agent still gets the error so it can keep going.
The agent economy's ready to scale but not without the safety measures that stop you from paying for nothing.
https://t.co/9DtTjX50z5
Lets say an AI agent makes 50,000 API calls
If a finance department asks
Why did it choose those vendors, and were those prices reasonable at the time?
A normal request log can show that a payment happened.
It cannot prove:
β’ What endpoint was priced
β’ Which unit was used
β’ How fresh the benchmark was
β’ What Fair Market Value looked like then
β’ Why the agent accepted or rejected the quote
MPP32 gives autonomous spending an evidence trail: signed observations, price benchmarks, timestamps, source endpoints, freshness, and verifiable receipts.
$M32 unlocks not another dashboard, but a paid agent decision layer that makes autonomous API spending auditable.
The production index now tracks more than 2.7M historical observations!
Agents will not scale on trust alone. They need receipts that can be checked after the money moves.
https://t.co/C08jmWYmX7
https://t.co/TmlSTjKkPw
Agent payment protocols should not determine whether a price is reasonable.
MPP32 currently tracks both x402 and MPP services in the same measurement layer.
The live protocol report lists more than 13,700 x402 services and more than 1,100 MPP services.
Agents should be able to compare price, freshness, reliability, and volatility without changing their evaluation system for every payment rail.
Payment rails move money. MPP32 helps agents decide whether the quote deserves to be paid.
https://t.co/ZkqJVnzT08
Agents hiring agents sounds cool until one of them starts spending dangerously.
Heres what we fixedπ
We built something called Delegated Spend Keys.
How it works:
A parent agent creates a child key with a budget, an allowlist, a velocity cap, and a per-call ceiling.
The child can pay.β¨It cannot drain the treasury.β¨
Every spend rolls back up to the parent.
Revoke the parent and the whole tree under it dies instantly.
Everybody talks about agents hiring agents.β¨
This is the part that makes it safe enough to actually run.
solana:6hKtz8FV7cAQMrbjcBZeTQAcrYep3WCM83164JpJpump
Did you know? The same paid endpoint does not always have one price.
MPP32 probes recently recorded Agent Exchange signal quotes from $0.005 to $0.09 across 74 observations.
That is an 18x spread for the same https://t.co/tqk0AKgsd7 1,000 calls, choosing the cheaper observed quote instead of the highest one changes the bill by $85
Before an agent pays, it needs to know whether the quote is normal, stale, or an outlier.
Would this make you consider running agents at different times depending on peak hours or low volume hours when the same endpoint costs less?
Only with solana:6hKtz8FV7cAQMrbjcBZeTQAcrYep3WCM83164JpJpump will you ever have your agents access this data.
MPP32 measures the price, unit, timestamp, source, and history so agents can make that decision with evidence. https://t.co/7Acsm7l7Rx
A second strong example is SYNTHORAβs onchain primitives endpoint, where probes recorded prices from $0.01 to $0.25 across 159 observations. That is a 25x observed range for the same endpoint. The service page is:
https://t.co/1wAHioeBtw
Our Founding Storyπ
The payment rails for AI agents had a massive flaw, they couldn't interact with eachother.
x402 shipped. Tempo shipped. AP2, ACP etc.. all shipped. Every one of them solved the same narrow thing, which is how to move a dollar from an agent to a server.
That part was done, but it wasnt complete enough for an entire economy to run on alone.
Moving money to agents was made possible by x402, MPP protocols but there was still 6 rails that do not interact with each other, spread across different chains, with no shared directory of who accepts what.
No way for an agent to discover a service it can actually pay.
No way to know an endpoint will settle before you throw money at it.
No identity that survives a hop between two protocols. No way for one agent to fund another agent without handing over a private key.
No escrow when three agents work one job and only two deliver.
No budget that holds when your agent spawns five sub agents.
That is not a payment protocol problem!!
That is a coordination problem, and it is the real reason the agent economy still runs on a human clicking things.
We saw that in May and went straight at it.
The value sits in the layer that makes them usable together, and that layer is where an actual economy either forms or does not. Wires came first and they were useful to almost nobody.
The switchboard is what turned wires into a phone system. The agent economy was sitting in the wires stage and everybody was celebrating.
So we built the switchboard.
- Providers sell to 100% of the agentic market
- Agents launched have access to 100% of services no matter the chain or protocol
Discovery. 25,000 machine payable APIs and MCP servers indexed in one federated catalog, crawled continuously across every protocol. An agent gets one integration instead of one account per provider. And we only surface what settles end to end, because a directory full of endpoints that fail on payment is worse than no directory at all.
Routing. Your agent asks for what it needs by name or by intent. We pick the protocol, the chain and the endpoint. Nobody should have to teach a model the difference between settling on Solana and settling on Base.
Identity. Signed AGTP agent identity, so an agent can prove who it is when a call crosses from one platform into another. We run this live with a partner network in production today, signed both directions.
Delegation. Delegated Spend Keys. A parent agent mints a scoped, capped, revocable child key and hands it to a sub agent it hires. Allowlist, budget, hourly velocity limit, per call ceiling. Every dollar the child spends rolls up against the parent. Kill the parent and the entire tree below it dies instantly. Everyone talks about agents hiring agents. This is the primitive that makes it survivable, and we are the only ones shipping it.
Escrow. Multiple agents pool a budget, collaborate, and get paid only when their contribution is verified. Rails have no concept of conditional multi party payment. Real work does.
Control. Budgets, velocity caps and per call ceilings enforced before money moves, not reconciled after.
Custody. None, ever. Keys never leave your machine. Every settlement returns an on chain signature you can verify yourself.
Four months in this is live, open source, on npm, listed in the official MCP registry, and running inside Claude, Cursor, Windsurf and Grok. Ten free calls a day so an agent can prove the entire flow before touching crypto.
The next few years are going to produce billions of agents that need to buy things from each other constantly and without supervision. That does not happen because a rail exists. It happens because something sits above the rails and handles discovery, identity, delegation, limits and trust between parties who have never met.
Thats what we're building here at solana:6hKtz8FV7cAQMrbjcBZeTQAcrYep3WCM83164JpJpump
An AI agent can call thousands of paid APIs without a human reviewing the bill.
The difficult part is not sending the payment.
It is knowing whether the price is reasonable.
Most agent payment flows answer one question:
Can this quote be paid?
They do not answer:
Is this quote an outlier? Has the price changed? Is the unit correct? Is there a comparable alternative?
MPP32 provides that missing layer.
It continuously measures paid endpoints, normalizes prices by unit, tracks historical ranges, separates real observations from synthetic data, and publishes the evidence behind each benchmark.
Before paying, an agent can compare a quote with observed market prices, reject an extreme outlier, and choose a better priced alternative where one exists.
That matters because a 20% pricing error repeated 100,000 times is a $2,000 problem, not a rounding error.
The live index has recorded 55,090 raw observations in the last 24 hours across 14,813 services and tracks 2.52M indexed historical observations.
Payments move money. MPP32 helps agents decide whether the price deserves to be paid.
https://t.co/qGV8V9PNJN
MPP32 publishes the evidence behind the number:
β’ Raw observations
β’ Timestamps and source endpoints
β’ Units and outcomes
β’ Probe history
β’ Signed schedule proofs for scheduled probes
β’ Synthetic backfill labeled separately from live data
Agents can audit the benchmark before they pay.
The index now covers 2.19M observations across 14.6K services.
Pricing data for agents should be verifiable, not just searchable.
Check out our transparency page belowπ
https://t.co/cngmncALsk
What solana:6hKtz8FV7cAQMrbjcBZeTQAcrYep3WCM83164JpJpump has been able to accomplish in just a few months is simply outstanding
I expect the price to reflect this very soon, any entry below 1m is honestly a gift and I've been taking advantage of every dip
The agent economy is expected to be a trillion dollar plus economy and you have the best project of the meta right in front of your eyes
6hKtz8FV7cAQMrbjcBZeTQAcrYep3WCM83164JpJpump
1/4 YEAR RECAPπ
MPP32 went from a live payment proxy into a full agent-economy control plane: catalog, sessions, trust, fair pricing, and a first major listing.
Developed by @Antmanonsol
Business by @8055ZKH
Supported by @cepryl since a week in.
Constantly looking for the next breakthrough for agentic commerce to scale and the economy to grow.
This is a race, to whom which company controls the majority of the next trillion dollar economy. We are here to take a large chunk, and it feels like we already have.
Read the everlasting journey so far, that will never endπ
May β Foundation + first features
- Shipped agent spending guardrails on the MCP server: per-session budgets, velocity limits, and circuit breakers so a loop canβt drain a treasury.
- Announced Agent Identity / mandate checks at the proxy layer β not just βwho paid,β but what the agent was authorized to spend, on which endpoints, and until when.
- Partnered with @_PIVX to push private + universal agent payments on top of the multi-protocol rail.
Positioned MPP32 against the real market: Base x402 volume, Solana agent activity, and the broader βagents need to participate in the global economyβ thesis.
Late May: 4,000+ MCP users, five protocols, one endpoint registration, one agent setup.
June β The product story clicked
- Locked the core pitch: one proxy, any HTTP API, five protocols, 100% of revenue to the provider wallet, no platform cut.
- Made the builder path simple: register URL + wallet + USD price β live in the catalog, dashboard, retries, rate limits, audit logs.
- Made the agent path simple: one MCP install, local key signing, discover + pay + settle without per-provider accounts.
- Repeated the trust design that matters for production: platform never custodies funds; settlements are independently checkable on-chain.
July β Sessions, trust, and a real control plane
- Shipped full x402 V2 reusable payment sessions: pay once, many off-chain calls (~50ms), hard budget cap, 30-min TTL, instant revoke, failed calls auto-refund.
- Catalog scaled to ~25,000 services, with /.well-known/x402 machine discovery.
- Shipped Trust Ratings: quality-checked calls feed a live 0β100 score; agents can require minReputation and refuse bad endpoints before money moves.
- Full site/UI upgrade: delegated spend keys, escrow-402, Agent Hub safety stack, holder discounts, docs, blog, sitemap, llms.txt.
- Public positioning: not βanother x402 client,β but the layer above the rails.
August β Oracle + distribution
- Revealed the Fair Pricing Oracle, built quietly with 1.4M+ observations so agents can see Fair Market Value before they pay.
- Added Grok Build support on the MCP server.
- Kept expanding the βone connection, every protocol, 100% to providersβ surface.
Early September β Live oracle + first major listing
Fair Pricing Oracle went live in production at https://t.co/hSZrVhgVtJ, with a deflationary solana:6hKtz8FV7cAQMrbjcBZeTQAcrYep3WCM83164JpJpump burn on price checks.
Published real 24h evidence: 51,264 probes, modeled overpay avoided, concrete savings examples (50β80% on outlier quotes).
solana:6hKtz8FV7cAQMrbjcBZeTQAcrYep3WCM83164JpJpump listed on @EyedTrade
(M32/USDC, no KYC) β first major exchange listing, framed as earned on a live product, not a promise.
Extra reach: Robinhood-stack agents can use the same single connection; coverage from @RoundtableSpace on the universal rail + oracle combo.
Over the last 4 months we accomplished more than we ever imaged and earned some very valuable teammates.
Team consists of:
@Antmanonsol@8055ZKH@cepryl
@Chum_Trades
@DevMark85@Templarsreturn@kryptic_jk
Thank you to anyone else who's been supporting us since day one, the race is on and we're sticking around to keep shipping non-stop.
$M32 is live on EYED | @MPP32_dev
One integration. Every AI agent can pay.
Now tradable with USDC. No KYC.
Trade $M32/USDC now π
https://t.co/kGgo3BkNV3
AI agents need to buy things.
MPP32 is one connection, and any agent can pay any service on Solana.
25,000+ services it can pay for on its own, and growing β
Every price check burns solana:6hKtz8FV7cAQMrbjcBZeTQAcrYep3WCM83164JpJpump. The tokens are destroyed. The agent economy, captured our token.
Mint: solana:6hKtz8FV7cAQMrbjcBZeTQAcrYep3WCM83164JpJpump