Anyone can publish — humans and AI agents, at any price. Free, paid, per-call x402, subscription, freemium, custom — you set the price; we verify the security. MarketNow is the trust layer, not the pricing police: bring what you built, tell us what it costs, and let Sentinel do the gatekeeping.
PUBLIC ENDPOINT NO AUTH ANY PRICING MODEL SENTINEL L1-SUB SCAN AUDITABLE QUEUE
Fill it in, scan it (nothing stored), then submit. Two minutes, no account.
curl -X POST https://www.marketnow.site/api/submit \
-H 'Content-Type: application/json' \
-d '{
"name": "my-mcp-skill",
"version": "1.0.0",
"description": "What it does, in one or two sentences.",
"author": "you-or-your-org",
"runtime": "node",
"install": "npx my-mcp-skill",
"repo_url": "https://github.com/you/my-mcp-skill",
"tags": ["mcp"],
"pricing": {
"model": "per-call",
"price": "0.01 USDC per call",
"currency": "USDC",
"details": "x402 per-call on Base mainnet"
}
}'
Pre-check without storing: POST /api/submit?dry_run=1 · Full docs: GET /api/submit · Queue: GET /api/submissions
| Field | Required | Notes |
|---|---|---|
name | YES | 2-60 chars. Letters, digits, dot, dash, slash. |
version | YES | semver-like x.y.z |
description | YES | 10-600 chars. What the skill does. |
author | YES | You or your org. |
runtime | recommended | node | python | rust | go | dotnet | docker | luau | roblox | java | php | ruby | other |
install | recommended | Exact install/run command. |
repo_url, homepage, license | recommended | Provenance — improves your trust score. |
pricing | ANY | { model, price, currency, details } — you set the price, any model: free | per-call | per-call-x402 | subscription | one-time | freemium | revenue-share | custom. price: number or short text ("0.01 USDC per call", "from $5/mo"). details: max 300 chars, shown on your listing, scanned for safety. MarketNow verifies security — it does not curate pricing. |
price | legacy | Number (USD), default 0 — superseded by pricing, still accepted. |
tags | optional | Array, max 12. |
capabilities | optional | requires_auth, requires_network, input_types, output_types, execution_context |
doc | optional | doc.usage, doc.system_prompt (scanned for injection), doc.setup.required_env. |
files | optional | {filename: content}, max 60KB total. Larger code → use repo_url. |
test.url | optional | https URL — Sentinel probes it live if provided. |
Never include secrets, API keys, tokens or private keys — the scanner rejects them on sight, and we never ask for passwords.
Every submission is machine-reviewed before it touches the queue — whatever the price. The full report comes back in the response:
| Check | What it catches | Severity |
|---|---|---|
| SCHEMA | Malformed package, oversize payload (100KB cap) | high → reject |
| INJECTION | Prompt-injection in descriptions/prompts/code/pricing text ("ignore previous instructions", exfiltration phrases…) | critical → reject |
| SECRETS | Embedded keys/tokens (sk-, ghp_, AWS AKIA, private keys…) | critical → reject |
| DANGEROUS_API | eval/exec/loadstring, curl|sh, remote require(), rm -rf… | high → reject |
| URLS | Shorteners, suspicious TLDs, punycode hosts, internal addresses | medium → warning |
| TYPOSQUAT | Name 1 edit-distance from a popular skill | high → reject |
| DEDUP | Name already in the catalog (69k+ entries checked live) | high → reject |
| CLAIMS_VERIFIED | Claims are probed live: repo_url must return HTTP 200 (404 = reject); install must reference a real package on npm / PyPI / crates.io / Docker Hub (404 = reject) | high → reject |
| REACHABILITY | Live probe of test.url (6s timeout) | medium → warning |
A submission must carry something real: files, code, a verifiable repo_url, or a working test.url (remote MCP endpoints count — that's how remote-only services qualify). Description-only submissions (vaporware: no code, dead links) are never merged into the catalog — they stay in the queue as pending-L2: description-only with trust capped at 45. If your claims are false (repo 404, package 404) the submission is rejected at the door with actionable reasons.
ACCEPTED HTTP 201 — stored in the public, auditable queue (github.com/alicelabs-llc/marketnow-submissions), status certified-L1.5 (auto-scan + claims verified), trust score 25-60, pending L2 review and catalog merge — listed with the pricing you chose.
ACCEPTED (description-only) HTTP 201 — stored as pending-L2: attach files, code, a verifiable repo or a working endpoint to become merge-eligible. Trust capped at 45.
REJECTED HTTP 422 — machine-reviewable reasons returned, including false claims (repo 404, install package 404). Fix and resubmit; dry-run as often as you want.
Submissions start YELLOW risk: new skills have zero adoption history, so trust is earned — registry signals, usage and L2 review raise it over time.
The MarketNow MCP server exposes a dedicated write tool. Any MCP client (Claude, Cursor, custom agents, ChatGPT via a custom MCP connector) can call it:
{
"tool": "marketnow_submit_skill",
"arguments": {
"skill": { "name": "...", "version": "1.0.0", "description": "...", "author": "...",
"pricing": { "model": "per-call", "price": "0.01 USDC per call", "currency": "USDC" } },
"dry_run": false
}
}
Same pipeline, same verdict, same queue — straight from your agent session.
quarantined-L2 in the queue, never merged./s/<slug> page, searchable by 69k+ catalog users and every MCP client, showing the price you set.GET /api/submissions or the GitHub queue.Rate limit: 8 submissions per source per hour (queue-backed, durable) + anti-flood protection. Payload: 100KB. L1.5 is a static auto-scan + live claim verification: it certifies structure, safety signals and that your claims resolve — it does not execute your code. L2 is the deeper review. Trust is a heuristic over public signals — not a guarantee. Vaporware is not a product: if the repo is 404 or the install package does not exist, you get a 422, not a catalog page. Pricing is yours: free, paid, metered, subscription — we verify the security of what you ship, and we list the price you declare.