DEV Community: 2x lazymac

MCP Tool Discovery Needs a Blocked-Channel Playbook

2x lazymac — Tue, 02 Jun 2026 05:01:46 +0000

This week I ended up writing down a rule that should have been obvious earlier:

If an MCP tool can only ship through one channel, it is not really ready to ship.

That sounds dramatic, but the operational pattern is simple. Discovery and distribution break more often than code does.

The real bottleneck was not the tool

Most small MCP tools are easy to explain:

what the tool does
what input it expects
what output it returns
where it helps inside Claude, Cursor, or other agents

The harder part is what happens after the implementation is done.

You still need users to find it.

In practice that means at least three separate systems have to work:

Marketplace or registry distribution
Package or deploy pipeline
Content distribution

When one of those channels is blocked, the others need to take over the same day.

Why content became the fastest fallback

A blocked dashboard can freeze a launch.

A short article usually cannot.

When marketplace auth breaks or redeploy is suspended, a dev.to post still does useful work:

it explains the use case
it creates search surface
it gives the tool a narrative
it buys time while the primary channel is repaired

That is why I now treat content as part of the shipping path, not as optional marketing.

The operating rule I use now

For any MCP tool or small API launch:

Ship one thing users can run
Ship one thing users can read
Keep one blocked-channel fallback ready

The fallback can be a second registry, an npm package, a direct install path, or a short technical article. The exact channel matters less than the fact that it already exists before the outage.

Discovery compounds only when the loop survives failure

Tool builders often focus on feature count:

more endpoints
more prompts
more adapters
more integrations

That work matters, but it does not compound if the distribution loop is fragile.

The better question is:

Can this tool still gain discovery today if the main launch path fails?

If the answer is no, the system still has a product gap.

Final takeaway

For MCP products, resilience is part of discovery.

A blocked marketplace should slow one channel, not stop the day's publish action. The product loop gets stronger when distribution is designed like infrastructure: with fallback paths, clear prechecks, and no dependence on a single happy path.

Social Media Post Analyzer API – Engagement Prediction, Sentiment, Virality Score

2x lazymac — Tue, 02 Jun 2026 04:06:30 +0000

Social Media Post Analyzer API

Analyze social media posts for engagement prediction, sentiment, virality score, and optimal posting time.

Quick Start

# Analyze a post
curl -X POST https://api.lazy-mac.com/social-post-analyzer-api/analyze   -H "Content-Type: application/json"   -d '{"text": "Just launched our new product\! Check it out.", "platform": "instagram"}'

Endpoints

Method	Path	Description
POST	/analyze	Analyze a post
POST	/predict-engagement	Predict engagement score
GET	/health	Health check

Features

Engagement prediction score
Sentiment analysis (positive/negative/neutral)
Virality potential indicator
Optimal posting time suggestions
Platform-aware analysis (Instagram, Threads, Twitter)

Deploy target: Cloudflare Workers
Live API: https://api.lazy-mac.com/social-post-analyzer-api/health

AI Lead Scoring API – B2B Lead Intelligence, ICP Match Score, Free Tier

2x lazymac — Tue, 02 Jun 2026 04:06:29 +0000

AI Lead Scoring API

Score B2B leads by fit, intent, and engagement. Returns ICP match score (0-100), conversion probability, and key signals.

Quick Start

# Score a lead
curl -X POST https://api.lazy-mac.com/ai-lead-scorer-api/score   -H "Content-Type: application/json"   -d '{"company": "Acme Corp", "industry": "SaaS", "employees": 150}'

Endpoints

Method	Path	Description
POST	/score	Score a single lead
POST	/batch	Batch score up to 100 leads
GET	/health	Health check

Features

ICP match score 0-100
Conversion probability estimate
Key signals: company size, industry fit, engagement indicators
Fast response under 500ms
Free tier available

Deploy target: Cloudflare Workers (global edge, <50ms p95)

GitHub: https://github.com/lazymac2x
Live API: https://api.lazy-mac.com/ai-lead-scorer-api/health

SaaS Subscription Cost Tracker API - REST + MCP, Free Tier

2x lazymac — Tue, 02 Jun 2026 04:05:30 +0000

SaaS Subscription Cost Tracker API

Track, normalize, and analyze SaaS subscription costs across tools. Detect unused subs, calculate ROI per seat, and alert on price changes.

Quick Start

# Health check
curl https://api.lazy-mac.com/subscription-cost-tracker-api/health

# API info
curl https://api.lazy-mac.com/subscription-cost-tracker-api

Endpoints

{'path': '/track', 'method': 'POST', 'description': 'Track a subscription'}
{'path': '/analyze', 'method': 'GET', 'description': 'Get cost breakdown analysis'}
{'path': '/health', 'method': 'GET', 'description': 'Health check'}

Pricing

Plan	Price	Features
Free	$0/mo	100 req/day, core endpoints
Pro	$Free / $5/mo	Unlimited, all endpoints, MCP

MCP Integration

This API supports the Model Context Protocol (JSON-RPC 2.0):

curl -X POST https://api.lazy-mac.com/subscription-cost-tracker-api/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","method":"tools/list","id":1}'

Get Your API Key

Get started on Gumroad — free tier available.

Built by lazymac | All APIs

PM Weekly Retro: Two Publish Rules We Locked This Week

2x lazymac — Mon, 01 Jun 2026 11:01:15 +0000

PM Weekly Retro: Two Publish Rules We Locked This Week

This week was not about shipping a big feature. It was about fixing a quiet failure mode: publish intention existed, but daily distribution still hit zero.

That sounds small until you zoom out. If a growth loop misses one day, the compounding loss is bigger than the missing post.

The Failure Pattern

We saw the same pattern repeat:

A queue existed in theory, but not always as an actionable file.
One distribution channel looked available, then failed on auth or policy.
The team assumed "we will publish later" instead of forcing a fallback path now.

In other words, the bottleneck was not content quality. It was operational ambiguity.

Rule 1: Minimum 2 Publish Actions Per Day

The first rule is simple:

Every day must contain at least 2 publish actions, not 1 intention.

Why two?

One action can fail because of auth, moderation, or platform drift.
Two actions force channel redundancy.
A daily floor is easier to audit than a vague weekly goal.

This changed the conversation from "did we have a good idea?" to "which two channels did we actually attempt today?"

Rule 2: Empty Queue Means Create New Content Immediately

The second rule fixed the more dangerous case.

If the publish queue has no usable candidate, the system must not wait for permission. It must create one new piece immediately.

That content does not need to be perfect. It only needs to be real, current, and distributable.

Good fallback topics this week:

PM weekly retro
KIS or market infra notes
MCP tool discovery
blocked-channel lessons that turned into policy

This sounds obvious, but operationally it matters because an empty queue is now treated as a trigger, not a pause.

What Actually Improved

Three things got clearer:

Channel health is now part of product ops.
Publish count is an execution metric, not a marketing vanity metric.
Fallback writing is cheaper than waiting for the "best" post.

That last point matters most. Waiting preserves comfort. Publishing preserves momentum.

The PM Takeaway

A surprising amount of growth work is not creative at all. It is just turning soft intentions into hard rules:

if queue missing -> inspect and recover
if channel blocked -> reroute
if no candidate -> write now

Once those rules exist, publishing stops depending on mood, memory, or optimism.

And that is the real win from this week: distribution became a system, not a hope.

When Your MCP Publish Channel Is Blocked, Content Becomes Infrastructure

2x lazymac — Fri, 29 May 2026 04:41:19 +0000

Today I hit a familiar operational pattern again:

The code path was fine. The publish path was not.

Our MCPize redeploy flow has been auth-blocked since April 21, 2026. That means the problem is no longer "fix the script." The problem is "design a system that still ships when one channel stays broken."

A blocked channel is not an exception anymore

Once a marketplace or registry is blocked for more than a day, it stops being a temporary incident.

It becomes part of the normal operating environment.

That changes the playbook:

Stop treating the primary channel as guaranteed
Precompute the fallback channels
Keep shipping through the fastest available route

For small developer tools, the fastest route is often not another dashboard.

It is a short piece of content.

Why content acts like infrastructure

When I say content, I do not mean a brand exercise.

I mean a distribution primitive that still works when deploy auth, marketplace approval, or package publishing is delayed.

A short technical post can still do useful work the same day:

explain what happened
document the fallback decision
create search surface
keep the project visible while other channels recover

That is infrastructure behavior. It preserves throughput under failure.

The publish order I trust now

For a new MCP tool or API, I now think in this order:

content for immediate distribution
registry when validation and auth are healthy
marketplace when the channel is actually available

This is not anti-marketplace. It is anti-single-point-of-failure.

If one blocked account can zero out your launch day, the product system is still too fragile.

The PM lesson

Most shipping delays do not start as roadmap problems.

They start as operational assumptions:

this token will still work
this dashboard will accept redeploy
this publish command will behave the same today

Good PM work for AI tools is partly about removing those assumptions before launch day.

Final takeaway

If your MCP tool depends on one publish channel, you are not done shipping.

You are waiting for that channel to decide whether your work is real.

A better system keeps at least one fallback path warm every day. For me, that increasingly means content first, then everything else.

MCP discovery test draft

2x lazymac — Fri, 29 May 2026 04:41:17 +0000

Header

Short body with enough text to pass minimum dev.to validation. Adding more sentences here.

ACWR 0.85–1.35 Sweet Spot 으로 러닝 부상 예측 — Translation Device 가 되어버린 코칭 앱

2x lazymac — Fri, 22 May 2026 12:14:39 +0000

요약 — 러닝 앱이 "더 빠르게"라고 외치는 동안 우리 몸은 매일 다른 신호를 보냅니다. RunVault 는 lnRMSSD 7일 이동평균, Acute:Chronic 부하 비율(ACWR), 좌우 보행 비대칭을 측정해 raw 수치 대신 보행 지침으로 번역 하는 Computational Sports Scientist 페르소나로 응답합니다.

왜 "더 빠르게 더 고통스럽게" 패러다임을 배격하는가

스트라바는 PR(Personal Record)을 강조합니다. 나이키 런 클럽은 오디오 가이드가 더 깊이 파고들라고 합니다. 런데이는 시작 전 광고를 보여줍니다. 어느 앱도 우리에게 가장 중요한 진실을 알려주지 않습니다.

"당신은 회복되지 않았습니다. 오늘은 달리지 마세요."

RunVault 는 이 격차를 메우려 만든 앱입니다. AI 코치는 두 정체성의 단일체 로 설계되었습니다.

정체성	역할
Computational Sports Scientist	HRV / 지상반력 / 비대칭 데이터의 엄격한 수치 분석
Empathetic Coach	과학적 결과물을 따뜻한 처방형 언어로 번역

이 이중 페르소나는 brand/GOVERNANCE_PROMPT_PROTOCOL.md 의 최상위 거버넌스로 LOCK 되어 있으며, 브랜드 바이블·색 토큰·LLM 시스템 프롬프트 위에 군림합니다.

1. HRV — 단일 측정값은 노이즈, 7일 평균이 신호

아침에 한 번 측정한 RMSSD 값은 통계적으로 거의 쓸모가 없습니다. 수면, 수분, 어젯밤 음주, 정신적 스트레스 — 이 모든 요인이 값을 흔듭니다. 그래서 Altini / Kiviniemi / Plews 의 선행 연구를 따라:

lnRMSSD = ln(RMSSD)
RMSSD_MEAN_7d = mean(lnRMSSD_1 .. lnRMSSD_7)
RMSSD_CV     = (std_7d / RMSSD_MEAN_7d) × 100

RMSSD_MEAN_7d 가 개인의 베이스라인 하한선 아래로 떨어지면서 동시에 RMSSD_CV 가 급감하면, Functional Overreaching (FOR) 로 판정하고 당일 트레이닝 계획을 자동으로 강도 다운그레이드합니다. 사용자의 의지력에 의존하지 않습니다.

2. ACWR Sweet Spot `[0.85, 1.35]` 가 친구입니다

Acute:Chronic Workload Ratio 는 프로 스포츠 (Catapult Vector, Zephyr BioHarness) 에서 검증된 부상 위험 지표입니다.

ACWR = AcuteLoad(최근 7일) / ChronicLoad(최근 28일 평균)

ACWR	상태	액션
< 0.85	낮은 부하	점진적 증량, 다음 주 +10-15%
0.85 – 1.35	Sweet Spot	지속
1.35 – 1.50	높은 부하 · 주의	다음 세션 강도 다운
≥ 1.50	위험 구간	강도 + 볼륨 강제 리셋

RunVault Cloudflare Worker 에서는 다음과 같이 구현됩니다:

const acwr = chronic > 0 ? acute / chronic : 0;
let status: 'low' | 'sweet_spot' | 'high' | 'danger' | 'no_data';
if (chronic <= 0) status = 'no_data';
else if (acwr < 0.85) status = 'low';
else if (acwr <= 1.35) status = 'sweet_spot';
else if (acwr < 1.5) status = 'high';
else status = 'danger';

UI 에서는 가로 게이지로 Sweet Spot 구간이 녹색으로 강조됩니다. raw 숫자 단독 노출은 절대 없으며, 모든 값에 텍스트 라벨이 동반됩니다.

3. 언어 번역 — raw 수치 단독 노출 금지

LLM 시스템 프롬프트 수준에서 금지 발화를 강제합니다:

const BUDDY_BANNED_PHRASES = [
  '부상입니다', '디스크입니다', '진단', '처방', '수술',
  '확실히 ', '반드시 ', '단언컨대',
  '더 빠르게', '더 고통스럽게', '한계까지', '죽을 만큼',
];

그리고 시스템 프롬프트는 이렇게 강제합니다:

== 언어 번역 (절대) ==
raw 수치 단독 노출 금지. 모든 수치 → "당장 취할 수 있는 보행 지침" 으로 번역.

예시)
  원시  → "오른발 충격이 왼발 대비 12% 높음"
  번역  → "오른발 착지 충격이 왼발 대비 12% 강합니다.
            우측 골반이 미세하게 뒤로 밀려 있을 수 있으니,
            상체를 곧게 펴고 명치를 가볍게 앞으로
            밀며 주행해 주세요."

여기가 Computational Sports Scientist 가 마이크를 Empathetic Coach 에게 넘겨주는 지점입니다.

4. 신뢰 스택 (Trust Stack) — 세 레이어를 항상 노출

건강 데이터 제품은 신뢰 위에서 부서집니다. 세 가지 UX 룰을 브랜드 거버넌스 최상단에 박았습니다:

Role Framing — "이 앱은 의료 장비가 아니라 번역 장치(Translation Device) 입니다." Onboarding + Settings 에 명시.
정확도 정직 공개 — 모든 카드 footer 에 "트렌드 기반. 단일 측정값은 신뢰하지 않습니다" 표시. 7일 이동평균이 최소 단위.
가시적 Data Autonomy — Privacy Center 가 설정 깊숙히 묻혀있지 않고 메인 홈 화면 일급 버튼. 폐기 + 로컬 격리 토글 즉시 접근.

Oura Ring 의 "Live Fast, Die Old" + Translation Device 프레이밍이 명시적 벤치마크입니다.

5. 아키텍처 — Perception → Network → Application 3-Layer

BoxingPro 시스템 아키텍처를 러닝 도메인에 처음 이식:

┌─ 1. PERCEPTION ─────────────────────┐
│ IMU + 인솔 압력 + PPG HRV           │
│ + 후방 카메라 MoveNet (17 키포인트) │
└──────────────┬──────────────────────┘
               ▼
┌─ 2. NETWORK ────────────────────────┐
│ 5차 버터워스 저역통과 필터          │
│ 타임스탬프 동기화 + 임계치 판정     │
└──────────────┬──────────────────────┘
               ▼
┌─ 3. APPLICATION ────────────────────┐
│ 거버넌스 프롬프트 시맨틱 매핑       │
│ LLM 코치 출력                       │
└─────────────────────────────────────┘

Application 레이어에서 LLM (현재 Claude Haiku + Pollinations 무료 fallback) 이 운동역학 데이터를 받아 금지어 스캐너를 통과시킨 후 한국어 코칭 문장을 출력합니다.

6. 5경쟁사 매트릭스 — 왜 차별화되는가

	NRC	스트라바	루나 (Runna)	런데이	RunVault
측정 데이터	페이스/심박	구간	마일리지	GPS	지상반력 / 제동력 / 좌우 대칭 / lnRMSSD / ACWR
AI 페르소나	직설적 코치	소셜 라이벌	플랜 생성기	한국어 페이서	Computational Scientist + Empathetic Coach
가격	무료 + 인앱	$11.99/월	$99.99/년	무료 + 광고	$5.99/월
약점	워치 단독 불안	영어 우선	프리미엄 가격	광고 과다	(초기) 센서 번들 진입 장벽

7. 직접 써보기

RunVault Closed Beta APK (Android, ~160 MB):
👉 https://api.lazy-mac.com/runvault.apk

앱을 열면 AI 코치 탭 이 첫 화면입니다 — 그래프 뷰 + 음성 마이크 + 프롬프트. "어제 무릎이 약간 아팠는데 오늘 10km 가도 될까?" 같은 질문을 던지면 페르소나가 작동합니다.

8. 다음

Health Connect HRV records SDK 실 wire (현재 Worker 는 D1 활동 데이터로 ACWR 만 계산; HRV 는 Phase F SDK 통합 대기 중 stub).
NURVV Run / Moticon 인솔 호환으로 실시간 보행 비대칭 검출.
Whoop Unite 스타일 B2B — 기업 웰니스 파일럿.

건강 데이터 제품을 만들고 있고 센서 숫자를 차트에 그냥 던지고 싶은 유혹이 들면, 제발 그러지 말아주세요. 번역 장치(Translation Device) 를 만드세요.

— Daniel (lazymac2x)

How I Built ACWR Sweet Spot Binding (0.85-1.35) Into a Running Coach — RunVault Translation Device

2x lazymac — Fri, 22 May 2026 12:00:49 +0000

TL;DR — I shipped a running app that refuses to say "run harder". Instead it computes Acute:Chronic Workload Ratio (ACWR), lnRMSSD 7-day moving average, and gait asymmetry — then translates those numbers into prescriptions like "shorten your stride by 5 cm so your foot lands under your hip." This post is the engineering + governance writeup.

The problem with "go faster, hurt more"

Open Strava and you see PRs. Open NRC and an audio coach tells you to dig deeper. Open Runday and you get an ad before your warmup. None of them tell you the most important truth your body is broadcasting through your heart rate variability:

"You haven't recovered. Don't run today."

That's the gap RunVault tries to close. We modeled the AI coach as two personas in one body:

Persona	Job
Computational Sports Scientist	Strict numeric analysis of HRV, GRF, asymmetry
Empathetic Coach	Translate scientific output into warm, prescriptive language

This duality is now governed by brand/GOVERNANCE_PROMPT_PROTOCOL.md — the document that sits above brand bible, color tokens, and even the LLM system prompt.

1. HRV — single point is noise, 7-day rolling is signal

Single morning RMSSD readings are statistically useless. Sleep, hydration, last night's wine, mental stress — all of them push the value around. So we follow the Altini / Kiviniemi / Plews protocol:

lnRMSSD = ln(RMSSD)
RMSSD_MEAN_7d = mean(lnRMSSD_1 .. lnRMSSD_7)
RMSSD_CV     = (std_7d / RMSSD_MEAN_7d) × 100

When RMSSD_MEAN_7d drops below the user's personal baseline and RMSSD_CV simultaneously collapses, we flag Functional Overreaching (FOR) and downgrade the day's plan automatically. No willpower required from the runner.

2. ACWR Sweet Spot — `[0.85, 1.35]` is your friend

The Acute:Chronic Workload Ratio comes from professional team sport (Catapult Vector, Zephyr BioHarness). It's defined as:

ACWR = AcuteLoad(last 7 days) / ChronicLoad(28-day average)

ACWR	Status	Action
< 0.85	Low volume	Gentle build, +10-15% next week
0.85 – 1.35	Sweet Spot	Sustain
1.35 – 1.50	High caution	Taper next session
≥ 1.50	Danger	Reset volume + intensity

In the RunVault Worker (Cloudflare Workers + D1) it's literally:

const acwr = chronic > 0 ? acute / chronic : 0;
let status: 'low' | 'sweet_spot' | 'high' | 'danger' | 'no_data';
if (chronic <= 0) status = 'no_data';
else if (acwr < 0.85) status = 'low';
else if (acwr <= 1.35) status = 'sweet_spot';
else if (acwr < 1.5) status = 'high';
else status = 'danger';

And it surfaces in the UI as a horizontal gauge with the sweet-spot band painted green. No raw number stands alone — every value gets a verbal label.

3. Language translation — raw numbers are banned

We literally banned a list of phrases at the LLM system prompt level:

const BUDDY_BANNED_PHRASES = [
  '부상입니다', '디스크입니다', '진단', '처방', '수술',
  '확실히 ', '반드시 ', '단언컨대',
  '더 빠르게', '더 고통스럽게', '한계까지', '죽을 만큼',
];

And the system prompt enforces:

== Language translation (mandatory) ==
Raw number-only output is forbidden. Every number must be translated
into "an actionable gait directive the runner can perform right now".

Example:
  Raw  → "Right foot impact 12% higher than left"
  Translated → "Right foot impact is 12% stronger.
                Your right hip may be slightly behind.
                Straighten your upper body and gently push
                your sternum forward as you run."

This is the part where the Computational Sports Scientist hands the mic to the Empathetic Coach.

4. Trust Stack — three layers, always visible

Privacy/health-data products fall apart on trust. Three UX rules now live at the top of the brand governance doc:

Role Framing — "This is not a medical device, it's a translation device." Surfaced in onboarding + Settings.
Honest accuracy disclosure — Every card footer says "Trend-based. Single measurements are not trusted." 7-day rolling is the smallest unit we report on.
Visible Data Autonomy — Privacy Center is on the main home screen, not buried in settings. Delete + on-device-processing toggles are first-class buttons.

The Oura playbook ("Live Fast, Die Old" + Translation Device framing) is the explicit benchmark.

5. Architecture — 3-layer Perception → Network → Application

Borrowed from BoxingPro and ported to running:

┌─ 1. PERCEPTION ─────────────────────┐
│ IMU + insole pressure + PPG HRV     │
│ + rear-cam MoveNet (17 keypoints)   │
└──────────────┬──────────────────────┘
               ▼
┌─ 2. NETWORK ────────────────────────┐
│ 5th-order Butterworth LPF           │
│ Timestamp sync + threshold judging  │
└──────────────┬──────────────────────┘
               ▼
┌─ 3. APPLICATION ────────────────────┐
│ Governance-prompt semantic mapping  │
│ LLM coach output                    │
└─────────────────────────────────────┘

The application layer is where the LLM (currently Claude Haiku with a Pollinations free fallback) takes the kinetic data, walks through the banned-phrase scanner, and emits Korean coaching prose.

6. Why this is a competitive moat

Five running apps benchmarked. Same table from brand/USP_v3.md:

	NRC	Strava	Runna	Runday	RunVault
Tracks	Pace/HR	Segments	Mileage	GPS	GRF / Braking / Symmetry / lnRMSSD / ACWR
Persona	Direct coach	Social rival	Plan generator	Korean pacer	Computational Scientist + Empathetic Coach
Price	Free + IAP	$11.99/mo	$99.99/yr	Free + ads	$5.99/mo
Weakness	Watch detached	English-first	Premium price	Ad clutter	(early) sensor bundle barrier

7. Try it

RunVault Closed Beta APK (Android, ~160 MB):
👉 https://api.lazy-mac.com/runvault.apk

The AI coach Tab is the first thing you see when you open the app — graph view + voice mic + prompt. Ask it something like "어제 무릎이 약간 아팠는데 오늘 10km 가도 될까?" ("My knee hurt a bit yesterday, can I do 10k today?") and watch the persona kick in.

8. What's next

Health Connect HRV records SDK wire (today the worker computes ACWR from real D1 activities; HRV is still a stub waiting on Phase F SDK integration).
NURVV Run / Moticon insole compatibility for real gait asymmetry detection.
Whoop Unite-style B2B integration for corporate wellness pilots.

If you're building a health-data product and tempted to just dump sensor numbers on a chart, please don't. Build a translation device.

— Daniel (lazymac2x)

I built a crypto signals API because the $300/mo bills made me angry

2x lazymac — Thu, 21 May 2026 13:52:29 +0000

I run a small API hub (api.lazy-mac.com). Last spring I needed crypto signals — buy/sell/hold calls with RSI, MACD, Bollinger — for a side project that experiments with rule-based scalping.

I tried three things:

CoinGecko Pro / CryptoCompare — fine for raw prices, but signal logic is on me. Cheapest plan with WebSocket was around $129/mo. The next jump for "give me an actual trade signal" comes from altFINS-style providers — currently $299/mo for their Standard plan.
Self-host TA-Lib + Binance — works for a day, then breaks on the third candle gap and you start writing your own retry/backfill instead of trading.
A small worker that pulls Binance klines, computes 6 indicators with explicit weights, returns one JSON per call.

Option 3 became crypto-signal-api. It's been my #2 most-called worker for months. So I cleaned it up and shipped the same thing for everyone else.

This post is what it actually does. With live data — copy-paste the curls and you'll get the same exact JSON I'm reading right now.

The actual ask

For trading signals you need three things in one response:

The current indicator block (RSI, MACD, EMA cross, Bollinger Band position, StochRSI, volume ratio)
A composite signal (STRONG_BUY / BUY / HOLD / SELL / STRONG_SELL) with a transparent score breakdown
A screener that ranks pairs so you can find what's actually moving today

Most providers give you one or two. So you stitch them. So you get bugs.

This worker does all three behind a single /signal/{symbol} endpoint, plus a /screener for ranking.

Try it right now

No API key. No signup. You can hit these as you read:

# One pair — full signal + indicators
curl https://api.lazy-mac.com/crypto-signal/api/v1/signal/BTCUSDT

Sample response, captured a few minutes before I published this:

{
  "symbol": "BTCUSDT",
  "interval": "1h",
  "signal": {
    "action": "STRONG_BUY",
    "strength": "strong",
    "confidence": 42,
    "score": 5,
    "maxScore": 12,
    "details": [
      { "name": "RSI",     "score":  1, "reason": "RSI low (36.78)" },
      { "name": "MACD",    "score": -1, "reason": "MACD bearish (hist: -123.43)" },
      { "name": "EMA",     "score":  0, "reason": "EMA mixed" },
      { "name": "BB",      "score":  2, "reason": "Price at lower Bollinger Band" },
      { "name": "StochRSI","score":  2, "reason": "StochRSI oversold (K:3.31 D:2.21)" },
      { "name": "Volume",  "score":  1, "reason": "High volume (1.6x avg)" }
    ]
  },
  "indicators": {
    "price": 76856.07,
    "rsi": 36.78,
    "macd": { "MACD": -38.15, "signal": 85.28, "histogram": -123.43 },
    "ema":  { "ema9": 77385.82, "ema21": 77499.17, "ema50": 77416.23 },
    "bollingerBands": { "upper": 79412.5, "middle": 78140.2, "lower": 76867.9 }
  }
}

Two things to notice:

The action is composite. Six sub-signals voted. RSI/BB/StochRSI/Volume pulled it bullish, MACD pulled bearish, EMA stayed neutral. Net +5 out of ±12. That's the entire scoring surface — no hidden ML, no proprietary "smart money" sauce.
Every component reports reason. If you're paper-trading and the signal disagrees with you, you can read why in one line.

Screener — what's worth looking at today

curl https://api.lazy-mac.com/crypto-signal/api/v1/screener

Returns the top 20 pairs ranked by signal score, with price, change24h, volume24h, signal, confidence. This is the endpoint I leave open in a tab.

Right now, at the top: BTCUSDT and XRPUSDT both STRONG_BUY with confidence 42, then ETHUSDT and ZECUSDT at BUY moderate. ZEC is up 11.3% on 24h with above-average volume — that's the kind of row you want to know about before scrolling Twitter.

Just the numbers, no opinion

If you only want the indicator block and want to roll your own scoring:

curl https://api.lazy-mac.com/crypto-signal/api/v1/indicators/ETHUSDT

{
  "symbol": "ETHUSDT",
  "indicators": {
    "price": 2110.79,
    "rsi": 37.76,
    "macd": { "MACD": -3.35, "signal": 0.11, "histogram": -3.46 },
    "ema":  { "ema9": 2122.72, "ema21": 2127.9, "ema50": 2129.7 },
    "bollingerBands": { "upper": 2152.25, "middle": 2131.77, "lower": 2111.29 },
    "stochRsi": { "k": 4.96, "d": 3.73 },
    "atr": 12.34,
    "volume": { "current": 14107.02, "average": 7790.6, "ratio": 1.81 }
  }
}

The Korean market thing

There's a second worker on the same hub that I keep underselling — korea-market. It's the only API I know of that wraps KOSPI/KOSDAQ prices, KRW exchange rates, and the Kimchi Premium (Korean exchange BTC price vs Binance, the gap retail traders watch for sentiment) in one MCP-ready endpoint.

curl https://api.lazy-mac.com/korea-market/api/v1/exchange

If you're building any AI agent that needs to know what Korean retail is doing on a given day, this is the lane. As of writing the Kimchi Premium just inverted to roughly -2%, meaning BTC trades cheaper on Upbit than on Binance — a sentiment signal worth catching in code, not by hand.

It's free.

Why the price is $14.99/mo and not $0 or $300

Three reasons I'll be honest about:

$0 forever doesn't last. Binance rate-limits aggressively. Caching at the edge is what makes this fast and not banned. Cloudflare Workers cost money. Free is a marketing tier, not a viable plan.
$300/mo is renting a brand. altFINS, Tokenmetrics — they have real teams and real datasets. They also have real overhead. If you want 150 indicators and human research notes, pay them. If you want six indicators with explicit math and an MCP endpoint, that's me at $14.99.
The price is what makes me ship. When someone is paying $14.99 for uptime, I show up. When it's free, I drift. That's the trade.

What you get on `/?q=crypto`

The whole hub is at api.lazy-mac.com/?q=crypto#catalog — three products visible there today:

Product	What it does	Tier
crypto-signal	RSI/MACD/Bollinger/SMA combined, 100+ pairs, screener, MCP endpoint	$14.99/mo
crypto-tax	FIFO/LIFO/Average cost basis. US/UK/EU tax brackets. CSV import.	$14.99/mo
korea-market	KOSPI/KOSDAQ + Kimchi Premium + KRW rates + news, MCP for agents	Free

The catalog page is live data — every count and request stat refreshes from the actual Cloudflare GraphQL feed, not a hardcoded number I wrote three months ago and forgot.

MCP — yes, it works in Claude/Cursor

Every product on the hub is also an MCP JSON-RPC 2.0 server. So you can give Claude or Cursor a literal "tell me which alt has the strongest signal right now" prompt and the model will hit the screener tool. The MCP endpoint is POST /crypto-signal/mcp if you want to wire it manually.

Where I'd love feedback

Three things I'm actively iterating on, comments very welcome:

Custom weights — should /signal/{symbol} accept a ?weights= query so traders can tune their own mix? My instinct is yes, but it explodes the cache space.
More exchanges — currently Binance only. Coinbase + OKX are next, in that order. Anyone particularly needs Bybit?
More indicators — Ichimoku Cloud is the most-requested. VWAP. Heikin Ashi candles. What would you actually use in production?

If you ship something with this — even a hobby paper-trading dashboard — drop a link below. I read every reply.

— Daniel
api.lazy-mac.com/?q=crypto#catalog

PM Weekly Retro: Three Publish Failures We Turned Into Rules

2x lazymac — Thu, 21 May 2026 11:01:59 +0000

PM Weekly Retro: Three Publish Failures We Turned Into Rules

This week in our AI product factory, shipping was not blocked by code quality. It was blocked by distribution friction.

We hit three concrete failures while launching new APIs and MCP tools:

1. MCPize Was Still Auth-Blocked

Our MCPize redeploy path has been blocked since April 21, 2026. That means "just redeploy everything" is no longer a reliable publish path.

Rule we adopted:

Treat marketplace lockout as a permanent operational risk
Keep alternative publish channels ready: npm, direct MCP registry, Smithery, dev.to
Fail fast and move traffic to channels we control

2. npm Token State Was Not Observable Enough

Two MCP packages were ready for release, but publish status depended on token freshness. The main lesson was simple: authentication is part of CI, not an afterthought.

Rule we adopted:

Run npm whoami as a lightweight publish precheck
Alert before token expiry becomes a launch blocker
Keep "ready to publish" packages separated from "ready to deploy" workers

3. Gumroad Error Handling Hid the Real Problem

One product create flow failed with an "unknown" error. The real issue was a long description payload hitting API constraints.

Rule we adopted:

Cap description length before submission
Log raw API response bodies on failure
Turn silent commercial failures into explicit engineering signals

What Actually Worked

The fastest reliable distribution channel this week was content.

When platform auth breaks or marketplace tooling stalls, writing a short operational post can still create discovery, trust, and inbound interest. A simple article moves faster than a blocked dashboard.

The Broader PM Lesson

PM for AI tools is not only roadmap and features. It is channel resilience.

If one launch path breaks, the team should already know the next two paths:

Content distribution
Package distribution
Registry and marketplace distribution

That order matters because speed matters.

Current Live Focus

We are continuing to ship small, production-ready tools around:

PII redaction and validation
LLM determinism checks
Prompt anti-pattern detection
Citation coverage analysis for RAG answers

The main operating principle is straightforward: fewer excuses, more channels, tighter feedback loops.

If you're building AI developer tools, measure your distribution system with the same rigor you use for product quality.

MCP Tool Discovery When a Publish Channel Is Blocked

2x lazymac — Thu, 21 May 2026 11:01:58 +0000

MCP Tool Discovery When a Publish Channel Is Blocked

This week the useful lesson was not about building another MCP server.

It was about what to do when one distribution channel stops cooperating.

Our MCPize redeploy path has been auth-blocked since April 21, 2026. That means the normal "ship code, push marketplace, move on" loop is no longer reliable. So the question changed:

How do you keep MCP tool discovery moving when one publish lane is dead?

1. Separate discovery from deployment

A lot of teams bundle these together:

deployment succeeds
therefore discovery happens

That is false.

Deployment puts the artifact somewhere. Discovery makes somebody care.

For small MCP tools, discovery still comes from a few simple surfaces:

short dev posts
npm installability
direct MCP endpoint docs
marketplace listings when they are available

If one of those surfaces breaks, the others should keep moving.

2. Small tools need small explanations

The best MCP tools I shipped recently were easy to explain in one paragraph:

what friction they remove
where they sit in the agent loop
why they should exist as a separate server

That rule matters because discovery quality is mostly explanation quality.

If a tool needs a long architecture defense before anyone understands it, it is probably not packaged tightly enough yet.

3. Blocked channels should trigger rerouting, not waiting

The operational pattern that kept working was simple:

If marketplace deploy works, use it.
If marketplace deploy is blocked, publish content the same day.
If content is already queued, push package distribution next.

That prevents a single vendor bottleneck from turning into a silent no-ship day.

4. MCP discovery improves when the tool catalog is opinionated

A large MCP catalog is not automatically helpful.

Discovery gets easier when each tool has:

one clear job
one install path
one quick example
one reason it beats hand-rolled prompt glue

That is also why tiny utilities keep outperforming abstract "agent platform" ideas in early distribution. They are easier to test, easier to describe, and easier to recommend.

5. Current playbook

The playbook I trust right now looks like this:

keep the publish loop daily
treat channel lockout as normal, not exceptional
write the shortest useful explanation first
prefer tools that fit into an existing agent workflow in under five minutes

That is less glamorous than a big platform launch, but it compounds faster.

Final takeaway

MCP product work is partly a tooling problem and partly a routing problem.

When one route is blocked, the system should not pause. It should reroute attention to the next surface that still works.

For us this week, that surface was content first and package distribution second.

DEV Community: 2x lazymac

MCP Tool Discovery Needs a Blocked-Channel Playbook

The real bottleneck was not the tool

Why content became the fastest fallback

The operating rule I use now

Discovery compounds only when the loop survives failure

Final takeaway

Social Media Post Analyzer API – Engagement Prediction, Sentiment, Virality Score

Social Media Post Analyzer API

Quick Start

Endpoints

Features

AI Lead Scoring API – B2B Lead Intelligence, ICP Match Score, Free Tier

AI Lead Scoring API

Quick Start

Endpoints

Features

SaaS Subscription Cost Tracker API - REST + MCP, Free Tier

SaaS Subscription Cost Tracker API

Quick Start

Endpoints

Pricing

MCP Integration

Get Your API Key

PM Weekly Retro: Two Publish Rules We Locked This Week

PM Weekly Retro: Two Publish Rules We Locked This Week

The Failure Pattern

Rule 1: Minimum 2 Publish Actions Per Day

Rule 2: Empty Queue Means Create New Content Immediately

What Actually Improved

The PM Takeaway

When Your MCP Publish Channel Is Blocked, Content Becomes Infrastructure

A blocked channel is not an exception anymore

Why content acts like infrastructure

The publish order I trust now

The PM lesson

Final takeaway

MCP discovery test draft

Header

ACWR 0.85–1.35 Sweet Spot 으로 러닝 부상 예측 — Translation Device 가 되어버린 코칭 앱

왜 "더 빠르게 더 고통스럽게" 패러다임을 배격하는가

1. HRV — 단일 측정값은 노이즈, 7일 평균이 신호

2. ACWR Sweet Spot [0.85, 1.35] 가 친구입니다

3. 언어 번역 — raw 수치 단독 노출 금지

4. 신뢰 스택 (Trust Stack) — 세 레이어를 항상 노출

5. 아키텍처 — Perception → Network → Application 3-Layer

6. 5경쟁사 매트릭스 — 왜 차별화되는가

7. 직접 써보기

8. 다음

How I Built ACWR Sweet Spot Binding (0.85-1.35) Into a Running Coach — RunVault Translation Device

The problem with "go faster, hurt more"

1. HRV — single point is noise, 7-day rolling is signal

2. ACWR Sweet Spot — [0.85, 1.35] is your friend

3. Language translation — raw numbers are banned

4. Trust Stack — three layers, always visible

5. Architecture — 3-layer Perception → Network → Application

6. Why this is a competitive moat

7. Try it

8. What's next

I built a crypto signals API because the $300/mo bills made me angry

The actual ask

Try it right now

Screener — what's worth looking at today

Just the numbers, no opinion

The Korean market thing

Why the price is $14.99/mo and not $0 or $300

What you get on /?q=crypto

MCP — yes, it works in Claude/Cursor

Where I'd love feedback

PM Weekly Retro: Three Publish Failures We Turned Into Rules

PM Weekly Retro: Three Publish Failures We Turned Into Rules

1. MCPize Was Still Auth-Blocked

2. npm Token State Was Not Observable Enough

3. Gumroad Error Handling Hid the Real Problem

What Actually Worked

The Broader PM Lesson

Current Live Focus

MCP Tool Discovery When a Publish Channel Is Blocked

MCP Tool Discovery When a Publish Channel Is Blocked

1. Separate discovery from deployment

2. ACWR Sweet Spot `[0.85, 1.35]` 가 친구입니다

2. ACWR Sweet Spot — `[0.85, 1.35]` is your friend

What you get on `/?q=crypto`