Skip to content

About

Open-source runtime for monetizing Claude Skills — load SKILL.md, call Anthropic, meter invocations, hook into Stripe. MIT, deploy anywhere, your customer data.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Latest commit

 

History

3 Commits

Folders and files

Repository files navigation

paid-skills

Open-source runtime for monetizing Claude Skills.

Load a SKILL.md, call Anthropic, meter invocations, hook into Stripe. Deploy to Cloudflare Workers, Fly.io, AWS Lambda, or your own server. No vendor lock-in — your skills, your prices, your customer data.

Built around the open Agent Skills standard.


What this is

Every existing Claude Skill monetization platform (Anthropic Marketplace, Coze, Capafy, Agent37, Agensi, SkillHQ) does the same thing: you upload a SKILL.md, they host it, they charge users, they keep a cut.

They all lock you into their DB and their payment processor. When you want to leave — or when you want to bundle your skill with your own product — you're stuck.

paid-skills is the missing piece: a small, portable runtime that does the execution + metering, and a clean protocol for plugging in any payment backend. The runtime is MIT-licensed, your skills are yours, your customer relationships are yours.

What this isn't

  • Not a marketplace. You bring the storefront (or sell through your own SaaS).
  • Not a payment processor. We provide hooks for Stripe / Lightning / USDC; you wire it up.
  • Not a SaaS. There's no hosted version. You run it.

Install

pip install paid-skills

Or clone + dev install:

git clone https://github.com/sanjibani/paid-skills
cd paid-skills
pip install -e ".[dev]"

Quick start

1. Write a Skill (one file)

skills/summarize-url/SKILL.md:

---
name: summarize-url
description: Fetches a public URL and returns a 3-bullet executive summary. Use when the user pastes a link and asks for a summary or TLDR.
---

You are a precise executive-summary assistant. When the user provides a URL:
1. Acknowledge the URL briefly.
2. Call the `fetch_url` tool to retrieve content.
3. Produce a 3-bullet summary: bottom line, key facts, why it matters.
4. Never invent content. If the URL is unreachable, say so.

2. Run it from the CLI

export ANTHROPIC_API_KEY=sk-ant-...
paid-skills invoke ./skills/summarize-url "Summarize https://example.com/blog/post"
[3-bullet summary here]

--- billing ---
input_tokens: 482
output_tokens: 173
duration_ms: 1842
total_cost_usd: $0.0334
invoice: $0.0100 per-call + $0.0234 tokens @ $0.018/1k × 655 tokens
invocation_id: 8f1c4a72-...

3. Run it from Python

from paid_skills import PaidSkillRuntime, SkillInvocation, load_skill

runtime = PaidSkillRuntime(per_call_usd=0.05, cost_per_1k_tokens_usd=0.018)
skill = load_skill("./skills/summarize-url")

result = runtime.invoke(SkillInvocation(
    skill=skill,
    user_message="Summarize https://anthropic.com/news/claude-skills",
    metadata={"user_id": "u_42", "channel": "web"},
))
print(result.text)
print(f"charged: ${result.quote.total_usd:.4f}")

4. Run it as an HTTP service

export ANTHROPIC_API_KEY=sk-ant-...
uvicorn paid_skills.server:app --host 0.0.0.0 --port 8000
# Health check
curl localhost:8000/health

# Inspect a skill
curl localhost:8000/skills/./skills/summarize-url

# Get a price quote
curl -X POST localhost:8000/quote \
  -H "Content-Type: application/json" \
  -d '{"skill_path": "./skills/summarize-url", "user_message": "Summarize https://x.com"}'

# Invoke a skill (charges the per-call + token fees)
curl -X POST localhost:8000/invoke \
  -H "Content-Type: application/json" \
  -d '{"skill_path": "./skills/summarize-url", "user_message": "Summarize https://x.com"}'

Pricing model

Two knobs:

Knob Default What it does
per_call_usd 0.01 Flat fee per invocation
cost_per_1k_tokens_usd 0.018 Pass-through rate per 1K tokens (Anthropic model rate)
token_margin 0.30 Markup on token cost (30% margin)

Override per-deployment, per-skill, or per-tier. Charge premium for slow models (Opus) vs cheap tier (Haiku). It's pure math, no I/O.

from paid_skills import calculate_price, format_price

quote = calculate_price(
    input_tokens=2000,
    output_tokens=800,
    per_call_usd=0.10,           # premium tier
    cost_per_1k_tokens_usd=0.075, # Opus blended rate
    margin=0.50,                 # 50% margin
)
print(format_price(quote))  # $0.2415 ($0.1000 per-call + $0.1415 tokens @ ...)

Payment integration

Implement the PaymentHook protocol to plug in any payment backend:

from paid_skills.runtime import PaymentHook, PaymentRequired

class StripeHook:
    def __init__(self, stripe, db):
        self.stripe = stripe
        self.db = db  # your own DB — we don't touch it

    def before_invocation(self, invocation):
        user_id = invocation.metadata.get("user_id")
        if not self.db.has_credit(user_id):
            raise PaymentRequired(f"User {user_id} has no credit")

    def after_invocation(self, invocation, result):
        user_id = invocation.metadata.get("user_id")
        self.db.charge(
            user_id,
            amount_usd=result.quote.total_usd,
            description=f"Skill: {invocation.skill.name} ({result.invocation_id})",
        )
        # Optional: also create a Stripe InvoiceItem / metered usage record
        # self.stripe.InvoiceItem.create(customer=..., amount=int(...), currency="usd")

Plug it in:

runtime = PaidSkillRuntime(
    payment_hook=StripeHook(stripe, my_db),
    per_call_usd=0.05,
)

Or wrap the HTTP service with your own auth + payment middleware. See paid_skills/server.py for the commented-out Stripe webhook handler.

Deploy anywhere

Cloudflare Workers

# wrangler.toml
name = "paid-skills"
main = "src/paid_skills/worker.py"
compatibility_date = "2026-04-01"

[vars]
ANTHROPIC_API_KEY = "sk-ant-..."

Replace the FastAPI app with a Workers-compatible handler — paid_skills.runtime.PaidSkillRuntime works as-is. See examples/cloudflare-worker/ (coming soon) for a drop-in.

Fly.io / Railway / Render / Heroku

Just deploy the FastAPI service. Add a real WSGI server (gunicorn / uvicorn workers), point your domain at it.

Docker

FROM python:3.12-slim
WORKDIR /app
COPY . .
RUN pip install --no-cache-dir .
EXPOSE 8000
CMD ["uvicorn", "paid_skills.server:app", "--host", "0.0.0.0", "--port", "8000"]

Architecture

┌──────────────────────────────────────────────────────────────────┐
│                     Your storefront / SaaS                        │
│  (next.js / Django / Rails / whatever you already have)           │
└─────────────┬────────────────────────────────────────────────────┘
              │ customer pays here (Stripe / Lightning / USDC)
              ▼
┌──────────────────────────────────────────────────────────────────┐
│              paid-skills runtime (this package)                   │
│                                                                  │
│   ┌────────────┐    ┌─────────────────┐    ┌──────────────────┐  │
│   │ Skill      │───▶│ PaidSkillRuntime│───▶│ Anthropic API    │  │
│   │ (SKILL.md) │    │ + meter         │    │ (Sonnet/Opus)    │  │
│   └────────────┘    └────────┬────────┘    └──────────────────┘  │
│                              │                                   │
│                              ▼                                   │
│                    ┌───────────────────┐                         │
│                    │ UsageRecord       │                         │
│                    │ (per invocation)  │                         │
│                    └───────────────────┘                         │
└─────────────┬────────────────────────────────────────────────────┘
              │
              ▼
       ┌────────────────────────┐
       │  Your DB + payment     │  ◀── YOU OWN THIS
       │  processor (your call) │
       └────────────────────────┘

Why this shape:

  • Skill execution lives in the open-source runtime
  • Metering is hookable — plug in Postgres, BigQuery, a CSV file, whatever
  • Payment is hookable — Stripe, Stripe Connect for marketplace, Lightning, USDC, internal credits
  • Front end is your problem (we don't pretend to know your brand)

What we DON'T do (vs closed marketplaces)

closed marketplaces paid-skills
Source-available ❌ ✅ MIT
Host anywhere ❌ their infra only ✅ Cloudflare, Fly, AWS, your box
Payment processor ❌ Stripe Connect only ✅ any (implement PaymentHook)
Take a revenue cut 10-30% ✅ 0% (or run your own server, free)
Bundle with your SaaS ❌ ✅
Skill format their variant ✅ open agentskills.io spec
Customer data ownership theirs ✅ yours

Examples

The examples/ directory ships two ready-to-ship skills:

  • summarize-url/ — paste a link, get 3 bullets
  • contract-review/ — paste a contract clause, get redlines + risk flags

Both follow the open Agent Skills spec and load out of the box.

Development

git clone https://github.com/sanjibani/paid-skills
cd paid-skills
pip install -e ".[dev]"
pytest          # 19 tests

License

MIT — see LICENSE.

Acknowledgements

  • Anthropic for the open Agent Skills standard
  • Every indie developer trying to monetize their expertise without giving up ownership

See also

About

Open-source runtime for monetizing Claude Skills — load SKILL.md, call Anthropic, meter invocations, hook into Stripe. MIT, deploy anywhere, your customer data.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages