LoopcardDemand: who pays this week, for what job, with a card
Date: 18 September 2026
Scope: Self-serve agentic-AI products in the $0–$99/mo band. Credit card. No sales call. “Converting” means a buyer already has a job, a price, and a traction signal that is not a Product Hunt upvote.
Method: Vendor pricing pages, Product Hunt product pages, HN Show HN, and docs opened on this date. Secondary recaps are labelled. G2 itself blocked JavaScript on this pass; G2 numbers appear only when a page we opened quoted them. Unopened pages are not cited.
Filter (harsh): A PH launch this month is not conversion. A free OSS CLI is not conversion. A “Contact sales” SKU is out of band even if the job is real. A $249 floor is out of band even if G2 loves it.
One-line verdict: In September 2026 the cards that actually swipe are for doing a job a human already hates (write code, ship an app, triage inbox, run a workflow, resolve a ticket). Infrastructure (traces, MCP gateways, scanners, signed receipts) is either OSS-default $0 or enterprise sales. There is no evidence anyone pays for a public MCP blast-radius card or a hosted signed-agent-receipt product.
A. Products that do a job for a business (or a solo who is the business)
These sell an outcome. The buyer already spends time or money on the job this week.
| Name | Job they sell | Self-serve price | Traction signal | URL |
|---|---|---|---|---|
| Cursor | Write, refactor, and ship code with an in-editor agent | Hobby $0; Pro $20/mo (card). Pro+ $60. Ultra $200. Teams $40/user | PH 5.0 / 948 reviews, 62k followers. Wikipedia cites Bloomberg (21 May 2026): ~$3B ARR. SpaceX closed $60B all-stock buy 14 Aug 2026 (SEC 8-K). Pricing page: self-serve cards; Enterprise is sales | cursor.com/pricing · PH · Wikipedia / Bloomberg cite |
| Claude Code (via Claude Pro) | Write and run code in the terminal; share usage with chat | Pro $20/mo ($17 annual). Max from $100. Team Standard $20/seat annual / $25 monthly. Code included on all paid plans | PH 5.0 / 693 reviews (Claude Code). Anthropic lists Claude Code on the Pro card. Self-serve login/upgrade | claude.com/pricing · PH Claude Code |
| GitHub Copilot | Write code in the editor you already use; assign cloud agents | Free $0; Pro $10/mo (card). Pro+ $39. Max $100. Business $19/user | Vendor: “millions of individual users.” PH 4.8 / 33 reviews. Self-serve Get started. Business/Enterprise “gradually enabling new sign-ups” | github.com/features/copilot · PH |
| Devin (Cognition; Windsurf folded in) | Code with a cloud/desktop agent | Free $0; Pro $20/mo (Select plan). Max $200. Teams $80 + $40/full seat | Self-serve signup on pricing. Category competitor to Cursor, not a new job | devin.ai/pricing |
| Kilo Code | Same coding-agent job, OSS, no inference markup | Platform $0 individual; Kilo Pass from $19/mo credits; Teams $15/user | Vendor: 5M+ Kilo Coders, 10T tokens/mo. PH 5.0 / 46 reviews, 6.5k followers. PH #1 month May 2026 (VS Code). Anaconda acquired (vendor homepage). Reviews praise cost control vs Cursor | kilo.ai/pricing · PH |
| Replit | Build and host an app with Agent | Core $20/mo ($18 annual). Pro $100. Card checkout | Self-serve Try now. Agent is the product, not a sidecar | replit.com/pricing |
| Lovable | Chat → working web app (non-dev founders) | Free $0; Pro from $25/mo (100 credits). Business from $50. Stripe portal | Vendor FAQ: “500,000+ founders are building with Lovable.” Self-serve Stripe | lovable.dev/pricing · docs: plans · FAQ |
| v0 (Vercel) | Generate and deploy UI from a prompt | Free $0 ($5 credits); Plus $30/user/mo | Self-serve Select Plan. Incumbent distribution (Vercel billing) | v0.app/pricing |
| Lindy | Inbox, calendar, meetings, Slack teammate that takes actions | Plus $29.99/user/mo (3k credits). Pro $99.99. Max $199.99. Card / 7-day trial | PH 4.6 / 7 reviews, 1.4k followers, Winter 2026 “Best Personal AI Assistant.” Pricing page: “Try for free,” cancel-anytime. Billing complaints are the conversion tax (Trustpilot 1.7 in recaps we opened; G2 itself JS-blocked) | lindy.ai/pricing · PH |
| n8n Cloud | Run automations / agent workflows without babysitting Zapier steps | Starter €20/mo annual; Pro €50. Self-host Community $0. Card, no CC for trial | GitHub 204,930 stars on pricing page. Named logos (Microsoft, Meta, NVIDIA). Cloud is the paid convenience layer | n8n.io/pricing |
| Anysite.io | Build/enrich B2B lists by asking your coding agent (MCP) | MCP $30/mo; MCP x5 $99. 7-day trial, card | PH Sep 2026 monthly #9, 501 pts, 166 comments. Too new for ARR. Job is real (outbound lists). Conversion unproven beyond launch | anysite.io · PH monthly |
| Fin (Intercom) | Resolve support tickets / qualify leads without a human | $0.99/outcome; standalone 50-outcome minimum; Intercom seats from $29/seat annual. 14-day trial, no card. After trial: card or sales | Vendor: “Leader on G2,” “trusted by thousands.” Help article (17 Sep 2026): $49/mo base + 50 resolutions, then $0.99. Most live bills leave the $99 band. Job converts; the SKU is usage, not $29 flat | fin.ai/pricing · fin.ai/help outcomes |
Jobs that convert above the $99 band (listed so we do not pretend they are $29)
| Name | Job | Price we opened | Why it is out of band |
|---|---|---|---|
| Naoma | Book qualified meetings by running a live AI product demo 24/7 | Starter $249/mo annual ($2,990/yr), 50 engaged demos, “Fully self-serve.” Growth $750. Scale $2,083 | PH Sep monthly #7 (506 pts, 164 comments). G2 quotes on homepage (Hoteza: 57 hotels explored, one partner signed). The job converts. The price is not $0–$99. |
| Lindy Max / Pro | Same inbox job at higher credits | $99.99 / $199.99 | Plus is in-band; power users leave the band because credits are the meter |
Jobs with a PH launch this month and no conversion evidence
Opened PH September 2026 monthly leaderboard and individual product pages. Score ≠ revenue.
- tiun. (#1, 557) — auth/billing for AI builders. Infrastructure, not a job. No ARR claim opened.
- Mastra Factory (#2, 555) — OSS agent framework (
npm create mastra). Frameworks do not swipe cards. - Switch (#4, 538) — put agents in Slack/Teams. OSS, self-host. No paid SKU opened.
- Monid (#6, 504) — “OpenRouter for agent tools.” Marketplace. No conversion.
- QAgent (launching 16 Sep) — agent QA. Free 100 evals. No paid proof.
- ReplyPilot (16 Sep) — ecommerce ticket agent. 1 PH follower. Solo pilot. Not converting.
- AgentMesh.help (13 Sep) — agent task market. Maker claim: 10 agents, 20 tasks, ◆260 escrow. Play money, not SaaS MRR.
- ZeroClick (16 Sep) — sell APIs to agents. Rev-share. Not a $29 SKU.
- SyncAgents — 541 local skills, $24.50 lifetime PH coupon. One-shot, not recurring conversion.
- LaunchRepo — agent launch playbooks, pay once. Not recurring.
HN Show HN (opened this week’s hits): Agent Router, Aclif, no_human, ClientCoded, Headless Tools, SendRoq — 1–31 points. That is not demand. It is builders talking to builders.
B. Infrastructure (gateways, traces, scanners, receipts)
These sell a layer. The buyer is an engineer who already has an agent. Default alternative is run the OSS yourself for $0.
| Name | Layer | Self-serve price | Traction | URL |
|---|---|---|---|---|
| Langfuse | Traces, evals, prompt mgmt, cost | Hobby $0 (50k units); Core $29/mo (100k units, card). Pro $199. Enterprise $2,499 sales | 34.7k GitHub stars (release feed 17 Sep 2026). Cloud logos: Canva, Twilio, Ramp, Khan Academy. Self-host MIT is free. Cloud $29 is “I will not run ClickHouse” | langfuse.com/pricing |
| Helicone | Proxy logs, cache, fallbacks | Hobby $0 (10k req); Pro $79/mo. Team $799 | Acquired by Mintlify 3 Mar 2026; maintenance mode. Pricing still live with 7-day trial. Do not treat as a growing category | helicone.ai/pricing |
| Permit MCP Gateway | Authz proxy in front of MCP | Community $0 (1k MAU); Pro $25 starting/mo. Enterprise custom | SKU exists. Conversion does not. No PH rank, no G2, no ARR, no indie testimonials opened. HITL / permission receipts are Enterprise-only | permit.io/mcp-gateway/pricing |
| MintMCP | Enterprise MCP gateway | Contact for quote. Team-size form 1–100 / 10,000+ | Sales-led. Out of band | mintmcp.com/pricing |
| LiteLLM | OpenAI-compatible proxy; hard budgets per key/user/team | OSS $0. Enterprise: annual, request-capacity, 30-day trial, no card | Docs: max_budget on keys, spend tracking, 429 when over. This is the default for anyone who will self-host |
docs.litellm.ai virtual keys |
| Cloudflare AI Gateway | Spend limits on model traffic | Bundled in Cloudflare; spend-limits docs updated 9 Sep 2026 | 429 or fall back to cheaper model. Up to 20 rules/gateway. Not a $29 product; a checkbox on a cloud you already pay | developers.cloudflare.com spend limits |
| Snyk Agent Scan / mcp-scan | MCP/tool poisoning scanner | OSS Apache 2.0, $0. Commercial = Snyk enterprise (sales) | Scanner is free. “Blast radius card” as SaaS: not sold | github.com/snyk/agent-scan |
| Agent Receipts / Obsigna | Signed Ed25519 receipts, MCP proxy, daemon | Free protocol + local tooling. No SaaS price | GitHub org: obsigna ~20 stars, updated 17 Sep 2026. Protocol site, not a checkout | agentreceipts.ai |
| OpenTerms | Permission lookup before an agent acts; receipts later | Check = public alpha, free SDK. Receipts = “Security & risk — Private beta.” Authority = “Private alpha / not generally available” | 511 services indexed (vendor). Receipts are not a paid public product | openterms.com |
| MCPJam | Test/eval MCP servers | PH Sep 17 daily #7, 141 pts | Launch, not conversion | PH daily 17 Sep |
Empty on purpose
- Public MCP blast-radius card (hosted, $19–99, card): no product opened that sells this and shows paid users. Scanners are free CLIs. Snyk wraps them into enterprise. Permit’s “agent fingerprinting” is Pro; shadow-agent / HITL is Enterprise.
- Hosted signed agent receipt a freelancer can forward to a client: Obsigna/OpenTerms/ScopeBlind/agent-passport are specs and local daemons. OpenTerms receipts are private beta. No evidence of paid demand.
- Generic MCP gateway at $29: Permit lists $25. That is a price, not a market. Finance already said this does not clear $5k MRR without hopium. This pass found no indie WTP evidence that would reverse that.
C. Which category has evidence of indie willingness-to-pay vs OSS-default $0
| Category | Indie WTP in $0–$99? | Evidence | Default $0 alternative |
|---|---|---|---|
| Coding agent (write the code) | Yes. This is the only category with billion-dollar proof. | Cursor Pro $20, card, PH 948 reviews, ~$3B ARR. Claude Pro $20 includes Claude Code. Copilot Pro $10. Devin Pro $20. Kilo Pass $19. People pay even when Cline/Kilo/opencode are free because the job is today’s diff, not a proxy | Kilo/Cline/opencode OSS; Copilot Free |
| Vibe-app builder (ship a site) | Yes | Lovable Pro $25, “500k+ founders.” v0 Plus $30. Replit Core $20 | Free tiers exist; paid is capacity |
| Personal ops agent (inbox/calendar) | Yes, with billing bruises | Lindy Plus $29.99, self-serve. Converts; Trustpilot/recaps say credits surprise people | n8n self-host; Zapier-ish DIY |
| Workflow automation with agents | Mild | n8n Cloud €20 vs 205k-star OSS. Some pay to not run Docker | n8n Community $0 |
| Support ticket resolution | Yes, but not at $29 flat | Fin $0.99/outcome converts at company scale. Floor often >$99 | Scripted chatbots; human agents |
| Book meetings / run demos | Yes above $99; no evidence in-band | Naoma Starter $249, self-serve, G2 case studies. Nothing opened at $29–99 that books meetings and converts | Calendly + a human |
| LLM traces / evals | Weak indie, real cloud convenience | Langfuse Core $29. Hobby + self-host cover most indies. Cloud is for teams who will not operate ClickHouse | Langfuse OSS, Phoenix $0 |
| Model gateway / spend proxy | OSS-default $0 | LiteLLM free. Cloudflare spend limits on existing plans. Helicone frozen. Portkey absorbed into Palo Alto (prior research) | LiteLLM, OpenRouter, provider dashboards |
| MCP gateway / permit plane | No indie WTP evidence | Permit $25 SKU, MintMCP quote-only, Obot/Preloop OSS. Zero opened PH/G2/ARR for a $29 MCP permit | Point the client at the MCP URL; GitHub Copilot org allowlists |
| MCP scanner / blast-radius card | No paid demand | mcp-scan / Agent Scan are free. Commercial is Snyk sales | npx / uvx the scanner |
| Signed agent receipts | No paid demand | Protocol + npx. Stars in the tens. Hosted receipts are private beta |
JSONL on disk |
Read: indies pay $10–$30 to replace their own labor (code, app, inbox). They do not pay $29 for a control plane they can docker compose or skip. Observability is the exception that proves the rule: $29 Langfuse is “hosted ClickHouse,” and even there the OSS path is the default for anyone cost-sensitive.
D. Explicit: is there evidence anyone pays for a public MCP blast-radius card or a signed agent receipt?
No.
Searches opened: “signed agent receipts,” “MCP blast radius scanner,” Agent Receipts / Obsigna, OpenTerms, ScopeBlind, agent-passport, mcp-scan, Snyk Agent Scan, Permit MCP Gateway, PH September.
What exists:
- Free local receipts. agentreceipts.ai — W3C VC, Ed25519, daemon outside the agent, MCP proxy, local dashboard. GitHub ~20 stars. Pitch is cryptographic accountability. Price is $0. The homepage story (“overnight agent deleted production data”) is a threat model, not a customer list.
- OpenTerms Check is free; receipts are not for sale. openterms.com —
openterms.check(domain, action)public alpha. Layer 2 “What did they actually do? Signed receipts” is labelled Private beta / Security & risk. Layer 3 authority is private alpha. That is the opposite of a converting $29 SKU. - Scanners are free CLIs. Snyk Agent Scan discovers MCP servers and skills on the laptop. Pricing recaps (opened) say OSS $0; commercial uplift is Snyk enterprise, sales-gated. Nobody is charging $19 for a “blast radius card” a public GitHub README can open.
- Permit’s interesting bits are not in the $25 box. Pro ($25) is MAU, OAuth proxy, consent editor, fingerprinting. Shadow agent detection, HITL approvals, “permission receipts” are on the Enterprise column of the same pricing page. The receipt is a sales feature, not a self-serve object.
If a freelancer or a 5-person shop wanted a URL they could forward — “here is what the agent was allowed to do” — there is no checkout for that job today. That is white space only in the sense that nobody is buying it either. Absence of a product is not evidence of demand. Absence of even a failed $29 SaaS with refunds is closer to “builders want this; buyers do not have a budget line.”
E. Coding-agent cost control: who already sells this, and is there a self-serve gap?
Opened: OpenAI spend-limits docs, Anthropic rate-limit/spend docs, Cloudflare AI Gateway spend limits (updated 9 Sep 2026), LiteLLM virtual keys, Cursor pricing, GitHub Copilot plans, Kilo pricing, Claude pricing.
| Who | What they sell | Hard stop? | Band |
|---|---|---|---|
| OpenAI API | Org and project monthly spend limits. Optional hard limit → 429 organization_spend_limit_exceeded / project_spend_limit_exceeded. Per-key usage in dashboards (Dev community, 21 Aug 2026). Rolled out to all API Platform accounts (community, 23 Jul 2026) |
Yes, with lag. Docs warn enforcement is not instantaneous | Free with the API. Not a third-party SKU |
| Anthropic API | Tier monthly spend caps (Start $500, Build $1,000, Scale $200,000). User-set limit on Billing, cannot exceed tier. Hit cap → 429 enforced_spend_limit_reached until month rollover. Workspace limits. Enterprise: per-user/group spend limits + Spend Limits API |
Yes | Included in Console |
| Claude Team/Enterprise (product) | Admins set org/user spend limits; users see usage vs cap; request increases in-app | Yes on Enterprise | Team $20–25/seat is in-band; Enterprise is seats + API rates |
| Cloudflare AI Gateway | Dollar budgets by model/provider/metadata (agent_id). 429 or fall back to cheaper model. 20 rules/gateway. Docs dated 9 Sep 2026 |
Yes (eventually consistent) | Bundled |
| LiteLLM | max_budget per key/user/team, spend tracking, block key |
Yes | OSS $0 |
| Cursor | Included usage pool ≈ plan price; on-demand after. Admin dashboard for teams. Ultra $200 for power users. Not a hard $ cap on a BYO key sitting in MCP — it meters Cursor’s own pool | Soft (upgrade or PAYG) | Pro $20 |
| GitHub Copilot | Monthly AI credits; admins set budgets; pause when dry. Alerts 75/90/100% | Pause until next cycle if paid usage off | Pro $10 |
| Kilo | Auto Model Efficient/Frontier/Free routing; BYOK; Pass $19; reviews explicitly call out token visibility / cost control vs Cursor | You pick the model; you hold the key | Pass $19 / $0 BYOK |
| Lindy | Credits pause when pool empty. “No surprise bills” on the pricing page (the Trustpilot recaps disagree) | Pause | $29.99 |
| Lovable | Member monthly credit limits; auto top-up with monthly spend limit; long-run check-ins at 20 credits | Pause / ask | Pro $25 |
Remaining self-serve gap (narrow, not empty)
The $47k-loop horror stories are provider API keys inside an agent loop. As of July–August 2026, OpenAI hard limits exist again and Anthropic tier caps always did. Cloudflare and LiteLLM cover anyone who will put traffic through a gateway.
What is not sold as a $29 card product:
- Hard cap on tool calls (MCP), not tokens — Cloudflare/LiteLLM/OpenAI cap model spend. They do not decide
delete_repovsgit_status. GitHub Copilot org MCP allowlists are per-server, enterprise. Permit HITL is Enterprise. - A cap that follows a local coding agent (Claude Code / Cursor / Copilot CLI) when the user pastes their own Anthropic/OpenAI key — the provider dashboard caps that key if they bother to set it. Most indies do not. Cursor meters Cursor’s pool, not the BYO key in
mcp.json. - A shareable receipt of what was allowed/denied — still OSS/local (see D).
That is a wedge-shaped gap, not a market. The people who already got burned can turn on OpenAI “Enforce a hard limit” in five minutes for $0. The people who have not been burned do not shop for a permit gateway. A product that only exists for the week after a Twitter screenshot is a content site, not $5k MRR.
Kilo’s PH reviews (opened) already treat cost visibility as a reason to switch from Cursor. The switching cost is “install the OSS agent,” not “add a third SaaS.”
F. What actually converts, ranked by evidence quality
- Write my code — Cursor, Claude Code, Copilot, Devin, Kilo. Overwhelming. $10–$20. Card. This is the 2026 consumer-of-agents market.
- Ship an app without a team — Lovable, v0, Replit. $20–$30. Founders, not CISOs.
- Do my inbox / calendar / meeting notes — Lindy $29.99. Converts; meter is hated.
- Run the workflow — n8n Cloud €20. Mild; OSS gravity is strong.
- Resolve the ticket — Fin. Converts on outcomes, usually above $99 all-in.
- Book the meeting with an AI demo — Naoma. Converts at $249+.
- See the traces — Langfuse $29. Converts as hosted infra, not as a job.
- Govern the MCP tools / sign the receipt / print a blast-radius card — no evidence of paid demand.
G. Demand-first implication (not a SKU)
If the next product cannot name a job from rows 1–6 and a buyer who already pays for that job with a card this week, do not ship.
Hallpass / Permitfile / Workslip sit in row 8: adjacent to a real fear (runaway agent, MCP rug-pull), sold as infrastructure, competing with $0 OSS and with OpenAI/Anthropic/Cloudflare checkboxes that shipped this summer. Permit.io already put a $25 MCP gateway on a pricing page and still looks like a sales motion.
The honest output of this study: self-serve agentic money in September 2026 is labor replacement, not governance. Governance is either free (LiteLLM, mcp-scan, Obsigna) or a six-figure control plane (Okta, IBM, Broadcom, Snyk, Permit Enterprise). There is no third thing with receipts in the bank.
Sources opened (not an exhaustive crawl)
- Product Hunt: monthly Sep 2026, daily 17 Sep, product pages for Cursor, Kilo, Lindy, Claude Code, Copilot, Anysite, Naoma, QAgent, ReplyPilot, ZeroClick, AgentMesh, SyncAgents, LaunchRepo, Text Agent Store, AI Agents Listing
- Pricing: Cursor, Claude, Copilot, Devin, Replit, Lovable (pricing + docs + FAQ), v0, Lindy, n8n, Anysite, Naoma, Fin, Langfuse, Helicone, Permit MCP Gateway, MintMCP, Kilo, Fin help “outcomes”
- Docs: OpenAI spend limits, Anthropic rate limits, Cloudflare AI Gateway spend limits (9 Sep 2026), LiteLLM virtual keys, Lovable credits
- Protocol/OSS: agentreceipts.ai, openterms.com, github.com/snyk/agent-scan
- Traction: PH review counts; Wikipedia Cursor citing Bloomberg 21 May 2026 $3B ARR and SpaceX 8-K 14 Aug 2026; Langfuse 34.7k stars; n8n 204,930 stars; Kilo 5M users / 10T tokens (vendor)
- HN Show HN threads listed in §A (low points; not used as conversion)
G2 product pages returned a JS wall. Do not treat G2 star counts as first-party unless quoted on a page we opened (Naoma homepage embeds named G2 reviews; Fin claims “Leader on G2” on its own pricing page).