研究内容为英文. 本页的研究内容（标题、简报、来源）以英文撰写，未经机器翻译。只有界面已翻译。

新想法Agent-services

Agent Cancel Desk

An MCP/API that personal AI agents call to actually cancel a user's subscription: it finds the official cancellation path, executes it with the user's explicit approval, and returns proof (protocol number or confirmation) — paid only on success.

3周前 · 评分于 2026年10月21日 · ID HBV-261021-08

暂定标题： Agent Cancel Desk (working name)

摘要

简报已验证自行申报假设

New / agent-services idea (hunch). Benchmark: flightclaw (agent books and pays for flights end-to-end and shows a visible service fee in its demo — Self-reported demo, $25.65 fee shown) and Agentik (pay-per-call tool catalog for agents; 2,429 tools, 302,049-impression launch post — Verified X).…

阅读其余部分

Neither cancels subscriptions. Existence check: Searches on 21 Oct 2026 found: Kordi MCP (returns a cancellation deep link/route, does not execute), LemSubs agent API (manages subscription data; its docs say it cannot cancel external vendor subscriptions), Recharge MCP (merchant-side cancellation for Recharge stores only), and a hackathon repo using voice calls (Kaeru). No agent-callable service that executes consumer cancellations across vendors and returns proof was found. X: 1,540 posts/7d (Down 12% Thu→Tue (full UTC days: Thu 15 251 → Tue 20 221); peak Thu 15 251; Tue 20 is 12% below peak; window 2026-10-14 10:00 UTC → 2026-10-21 10:00 UTC (Wed 14 Oct and Wed 21 Oct are partial UTC days; full days Thu 15 → Tue 20)). Revenue hypothesis: ~$2K–$8K/month at month 9 (HYPOTHESIS: 1,500–4,000 confirmed cancellations × ~$1.50–$2 net).

描述假设

阅读其余部分

sistants, OpenClaw-type personal agents) acting for people who want a subscription gone; human beneficiaries are adults in Brazil first, then the US/Mexico. Pain: Cancelling is the step agents still cannot finish: existing agent tools return a link or a route, and the human still has to log in, find the hidden button or call support. People complain publicly every day that they cannot cancel. Monetization: Success fee per confirmed cancellation paid from the agent's wallet (hypothesis $1–$3); free route lookup; listing on agent tool marketplaces (Agentik-type) as distribution. stack: MCP server + Next.js approval page on Vercel; Postgres route DB; x402 or prepaid wallet billing. GTM: List on MCP registries and agent tool catalogs; SEO/GEO pages "how to cancel <service> in Brazil" that agents and humans cite; faceless how-to clips.

目标客户

Primary user: consumer AI agents (ChatGPT, Claude, Meta Muse-style assistants, OpenClaw-type personal agents) acting for people who want a subscription gone; human beneficiaries are adults in Brazil first, then the US/Mexico.

痛点

Cancelling is the step agents still cannot finish: existing agent tools return a link or a route, and the human still has to log in, find the hidden button or call support. People complain publicly every day that they cannot cancel.

为何是这个分数

分数较低的原因： Vendors block automation or require phone calls; success rate may be too low to earn a per-success fee.

护城河假设

Route DB + success data: Fresh, tested cancellation routes per vendor and country are hard to keep current. Every success/failure improves routing; agent platforms prefer the tool that finishes the job. Status: Hypothesis.

需求信号

X 上的帖子（7 天）

1540

信号趋势

Down 12% Thu→Tue (full UTC days: Thu 15 251 → Tue 20 221); peak Thu 15 251; Tue 20 is 12% below peak

网络需求已验证

Cancellation pain: 1,540 posts/7d in EN+PT, flat (Verified X). Agent-capability pain: 1,172/7d (Verified X). Search volume for "how to cancel X" not measured this run. Other X counts (same window, Verified X API): 1,172 posts: ("for my agent" OR "agent needs" OR "agents need" OR "agent can't") (pay OR book OR buy OR login OR password OR phone) lang:en

市场与竞争对手

TAM

Not sized this run — Scope: Consumer subscription management. Method: No market report fetched. (Basis: n/a)

SAM假设

~$100K–$500K / yr (hypothesis) — Scope: Agent-initiated cancellations BR+US. Method: HBV assumption. (Basis: Hypothesis)

SOM假设

~$24K–$96K ARR (hypothesis) — Scope: Month-9 run-rate. Method: From rev ×12. (Basis: Low)

竞争对手

Kordi (routes), LemSubs (subscription data), Rocket Money and similar apps (human-facing concierge, not agent APIs), consumidor.gov.br (Brazil complaint channel). No direct agent-executing competitor found.

商业模式

变现方式假设

Success fee per confirmed cancellation paid from the agent's wallet (hypothesis $1–$3); free route lookup; listing on agent tool marketplaces (Agentik-type) as distribution.

收入假设（每月）假设

~$2K–$8K/month at month 9 (HYPOTHESIS: 1,500–4,000 confirmed cancellations × ~$1.50–$2 net).

待完成的任务

When my user told me to cancel a subscription, I want to execute the cancellation with one approval and get proof, so I can report back done instead of handing the user a link.

HXC

Agent of Paula, 38, Recife (illustrative) · Paula asks her assistant to "cut my subscriptions"; it can list them but stalls at cancelling. Buying signal: Her agent calls Cancel Desk, she taps approve, and gets a protocol number — the agent pays R$8 from its balance.

构建与 GTM

技术栈

MCP server + Next.js approval page on Vercel; Postgres route DB; x402 or prepaid wallet billing

GTM 渠道

List on MCP registries and agent tool catalogs; SEO/GEO pages "how to cancel <service> in Brazil" that agents and humans cite; faceless how-to clips

风险

主要风险

Vendors block automation or require phone calls; success rate may be too low to earn a per-success fee.

全部风险已验证

Value: Supported — Pain Verified daily | Usability: Attention — User must trust agent to act | Feasibility: High — Vendors resist automation | Business Viability: Unvalidated — Per-success WTP by agents

下一步验证

Build routes + execution for 20 top Brazilian subscriptions; kill if fewer than 60% of 100 test cancellations succeed end-to-end or fewer than 3 agent platforms list it in 30 days.

来源

kordiapp.com (via github.com/kordi-labs/kordi-mcp-server)
lemsubs.com
agentik.cc
X API counts

ideaagent-servicesmcpsubscriptionscancellationsuccess-feebrazilusbenchmark-flightclawbenchmark-agentik

想在评分当天收到下一个想法吗？

分数、来源、风险和下一步，通过邮件发送。

ProductBots.ai

喜欢这个想法吗？用 ProductBots.ai 把它变成计划。

ProductBots.ai 是一个 Agentic Product OS：由智能体与你一起完成产品工作，从发现到计划。下载这个想法的简报，作为起点。

打开 ProductBots.ai

简报文件为英文。

预览简报

# Briefing: Agent Cancel Desk (working name)

Source: Startup Ideas HQ (HB Ventures), found 2026-10-21. ProductBots Opportunity Score: 61/100, evidence confidence 45%.
Evidence labels (Verified, Self-reported, Hypothesis) are kept as written by the researcher.

## Goal for ProductBots.ai
Turn this idea into a validated MVP plan and a first build. Start from the next validation step; do not assume unlabelled numbers are verified.

## What it is
An MCP/API that personal AI agents call to actually cancel a user's subscription: it finds the official cancellation path, executes it with the user's explicit approval, and returns proof (protocol number or confirmation) — paid only on success.

New / agent-services idea (hunch). Benchmark: flightclaw (agent books and pays for flights end-to-end and shows a visible service fee in its demo — Self-reported demo, $25.65 fee shown) and Agentik (pay-per-call tool catalog for agents; 2,429 tools, 302,049-impression launch post — Verified X). Neither cancels subscriptions. Existence check: Searches on 21 Oct 2026 found: Kordi MCP (returns a cancellation deep link/route, does not execute), LemSubs agent API (manages subscription data; its docs say it cannot cancel external vendor subscriptions), Recharge MCP (merchant-side cancellation for Recharge stores only), and a hackathon repo using voice calls (Kaeru). No agent-callable service that executes consumer cancellations across vendors and returns proof was found. X: 1,540 posts/7d (Down 12% Thu→Tue (full UTC days: Thu 15 251 → Tue 20 221); peak Thu 15 251; Tue 20 is 12% below peak; window 2026-10-14 10:00 UTC → 2026-10-21 10:00 UTC (Wed 14 Oct and Wed 21 Oct are partial UTC days; full days Thu 15 → Tue 20)). Revenue hypothesis: ~$2K–$8K/month at month 9 (HYPOTHESIS: 1,500–4,000 confirmed cancellations × ~$1.50–$2 net).

## Customer and pain
Target customer: Primary user: consumer AI agents (ChatGPT, Claude, Meta Muse-style assistants, OpenClaw-type personal agents) acting for people who want a subscription gone; human beneficiaries are adults in Brazil first, then the US/Mexico.
Pain: Cancelling is the step agents still cannot finish: existing agent tools return a link or a route, and the human still has to log in, find the hidden button or call support. People complain publicly every day that they cannot cancel.

## Market and competitors
TAM: Not sized this run — Scope: Consumer subscription management. Method: No market report fetched. (Basis: n/a)
SAM: ~$100K–$500K / yr (hypothesis) — Scope: Agent-initiated cancellations BR+US. Method: HBV assumption. (Basis: Hypothesis)
SOM: ~$24K–$96K ARR (hypothesis) — Scope: Month-9 run-rate. Method: From rev ×12. (Basis: Low)
Competitors: Kordi (routes), LemSubs (subscription data), Rocket Money and similar apps (human-facing concierge, not agent APIs), consumidor.gov.br (Brazil complaint channel). No direct agent-executing competitor found.
Geography: Brazil first (PT), then US and Mexico

## Business model
Monetization: Success fee per confirmed cancellation paid from the agent's wallet (hypothesis $1–$3); free route lookup; listing on agent tool marketplaces (Agentik-type) as distribution.
Revenue hypothesis (monthly): ~$2K–$8K/month at month 9 (HYPOTHESIS: 1,500–4,000 confirmed cancellations × ~$1.50–$2 net).

## Build and go-to-market
Build effort (agent-weeks): unknown
Time to first revenue: ~3–5 weeks after launch
Tech stack: MCP server + Next.js approval page on Vercel; Postgres route DB; x402 or prepaid wallet billing
GTM channels: List on MCP registries and agent tool catalogs; SEO/GEO pages "how to cancel <service> in Brazil" that agents and humans cite; faceless how-to clips

## Risks and next step
Main risk: Vendors block automation or require phone calls; success rate may be too low to earn a per-success fee.
Next validation action: Build routes + execution for 20 top Brazilian subscriptions; kill if fewer than 60% of 100 test cancellations succeed end-to-end or fewer than 3 agent platforms list it in 30 days.

## Sources
https://kordiapp.com/mcp (via github.com/kordi-labs/kordi-mcp-server); https://lemsubs.com/agent-api; https://www.agentik.cc; X API counts

这是研究，不是投资建议。数字保留其发布时的标签。

摘要

简报✓已验证◐自行申报○假设

描述○假设