Business judgment on AI products
When an assistant needs to register an account, it orders a phone number in the right country, reads the SMS code, and no one has to wait.
01
Start inside the user's day · Public facts + observable behavior · 2026-08-27
When an assistant needs to register an account, it orders a phone number in the right country, reads the SMS code, and no one has to wait.
Public materials do not yet show how users complete this job today or what they replace.
The product targets friction in this job, but public user evidence does not yet show the cost, frequency, or consequence of leaving it unsolved.
An assistant that gets things done needs an identity, and the smallest unit is a phone number. The trend is human-facing APIs rewritten for assistants. The entry is the SMS-code step in cross-border signup, charged per use and discarded after.
Its public repository has 166 stars and 2 forks, showing developer attention; repeat use and payment are not yet verified.
① Whether stars pass 1,000 in three months — whether it still grows after MCP hype; cools is the real demand signal; ② Whether an agent platform (Claude Code, Cursor) builds it in or partners with it —; being embedded is both absorption and category validation; ③ Whether any ban/compliance incident …
Worth trying. Its public repository has 166 stars and 2 forks, showing developer attention; repeat use and payment are not yet verified.
if your product or API is repetitive, autonomous, and usage-priced, seriously consider an agent-native entry (MCP or skill) and price it as pay-per-result with automatic refunds on failure. Agent customers do not read your site and do not call support; they read your API docs and your failure rate. pay-per-call plus auto-refund on failure plus rate-limit-not-quota. This combination persuades agent customers far better than a subscription.
Pay per code, prepaid balance. Typical price $0.43 per verification (from $0.01);; if no code arrives within 20 minutes, the balance refunds automatically. Day-rental for; accounts that must keep receiving codes. 150 requests per second. No… ① Whether stars pass 1,000 in three months — whether it still grows after MCP hype; cools is the real demand signal; ② Whether an agent platform (Claude Code, Cursor) builds it in or partners with it —; being embedded is both absorption and category validation; ③ Whether any ban/compliance incident …
When an assistant needs to register an account, it orders a phone number in the right country, reads the SMS code, and no one has to wait.
Its public repository has 166 stars and 2 forks, showing developer attention; repeat use and payment are not yet verified.
An English validation note will follow from the public evidence.
The assessment is recorded; an English explanation is pending.
The assessment is recorded; an English explanation is pending.
The assessment is recorded; an English explanation is pending.
The assessment is recorded; an English explanation is pending.
02
Market comparison
Local supply: Emerging
Demand evidence: Not yet verified
Public coverage has been recorded for this market. · 2026-08-27
03
The call and next move come first; the full read retains the evidence and counterevidence.
An MCP server that hands AI agents a disposable phone number: when an agent needs to sign up somewhere, it orders a private number in the right country, waits for the SMS code, reads it, and returns the number — no human at the keyboard.
SV Number (sms-verification-number.com), a company already running a phone-verification
business, repackaged its existing SMS API as an MCP server (sv-number/mcp-server,
MIT, TypeScript) and aimed it squarely at AI agents. The site has a landing page built
for agents plus an agent-skill (plain markdown) for Claude Code and Codex.
Read: this is not a new product launch, it is a channel redo. Same SMS-verification service, but the user changed from a person to an agent. Nothing new was invented — the interface language was switched from a human-facing web page to MCP tools an agent can read. Enterprise DNA covered the pivot explicitly, treating it as the reference case for "rewrap a vertical API for agents."
country=<id>, and the agent picks based on
live availability (getCountryAndOperators, getServicesAndCostWithStatistics)
instead of a list pinned in its promptgetStatus returns STATUS_OK:123456
directly — the code arrives already parsed, no regex over SMS textlist_countries, list_services, order_number,
wait_for_code, finish_activation, cancel_activation, totp_code, etc.,
usable from Claude Code, Codex, and Cursor; without MCP, the underlying methods
(getNumber, getStatus, setStatus) are the market-standard names, so an existing
client mostly changes its base URLWhat it deliberately does not do: no SMS sending, no calls. The numbers receive verification codes and nothing else.
When an agent needed to complete a signup (say, an Indonesian-market account), the bottleneck was phone verification: verification services look at the number's country, and a US number registering an Indonesian account gets refused before the SMS is sent.
The old flow was a human at an SMS-activation platform (sms-activate and peers): choose a country, place an order, wait for the code, paste it back into the signup — minutes of human babysitting per account. Earlier still: buying a physical SIM and plugging it in.
SV Number automates all three steps: one API call orders, polls, and completes the signup, with the country decided by request parameter. It replaces the act of a person standing by waiting for a code. More valuable: it makes "a fresh number per account" the default — services remember which numbers already verified an account, so a permanent number fails on the second signup, while this market was always priced as one-code-per-activation anyway.
Pay per code, prepaid balance. Typical price $0.43 per verification (from $0.01); if no code arrives within 20 minutes, the balance refunds automatically. Day-rental for accounts that must keep receiving codes. 150 requests per second. No subscription, no seat fee.
Read: this pricing is designed for agent customers — agents do not negotiate, do not flinch at a few cents per call, and keep running as long as the API is stable. For the seller this is the ideal customer: usage grows by itself, near-zero service cost. The risk is on the compliance side: phone-verification inherently sits next to batch signups and fraud, which is the sword hanging over every provider in this category.
| Dimension | Call |
|---|---|
| Founder-product fit | An existing verification business turning itself around; the pain (agents need an identity) was fed to them by the market, not invented |
| Product insight | Nailed "the smallest unit of identity is a phone number"; delivering the parsed code as an MCP tool is closer to real demand than the crowd building "call/text for agents" |
| Execution quality | The MCP wrapper itself is low-tech, but the agent-skill, refund policy, comparison pages, and live-availability API make this a finished product, not a demo |
| Timing | Good. Agents signing up for accounts at scale is a growing need, and 2026 is MCP's channel-window |
This is the textbook case of "rewrap a vertical API for agents," worth studying by anyone running an API business. SV Number invented no new service. It did three small things: changed the interface language to what agents read, changed the docs to what an agent can read by itself, and changed the pricing to something an agent never flinches at. Together, three small things turned a human-facing niche business into agent-facing infrastructure.
The transferable rule: take any vertical service you already use and run it through this template — can an agent call it without waiting for a human, without reading docs, and pay per use automatically? First ask "can this service run autonomously." Only services that can should get their interface redone.
Three risks to watch: compliance — verification sits one line away from batch signups and fraud, and platform bans or legal action can hit at any time; it is a channel business — stars rose because the MCP ecosystem is rising, not because it is irreplaceable, and a big player (Twilio and peers) getting into MCP is only a matter of time; and agent-facing pricing invites faster price wars than consumer markets when competitors arrive.
① Whether stars pass 1,000 in three months — whether it still grows after MCP hype cools is the real demand signal ② Whether an agent platform (Claude Code, Cursor) builds it in or partners with it — being embedded is both absorption and category validation ③ Whether any ban/compliance incident becomes public — the life-or-death line for this kind of business
Product logic: if your product or API is repetitive, autonomous, and usage-priced, seriously consider an agent-native entry (MCP or skill) and price it as pay-per-result with automatic refunds on failure. Agent customers do not read your site and do not call support; they read your API docs and your failure rate.
Pricing structure: pay-per-call plus auto-refund on failure plus rate-limit-not-quota. This combination persuades agent customers far better than a subscription.
Worth watching. A clean channel pivot that turned "giving agents an identity" into a transparent, usage-priced business. But its moat is not technical — it is compliance handling and speed of response to the agent ecosystem, both unproven. Come back in three months against the three checks above.
04
Evidence trail
05
Use these searches when the official site is missing or the current link is only a lead.