MCP 服务器

@melis-ai/x402-tools-mcp

An MCP server providing 22 pay-per-call utility tools for AI agents (scrape, validate, embed, store, moderate, notify, convert, prevent loops) without accounts or API keys, using USDC payments via the x402 protocol.

README

@melis-ai/x402-tools-mcp

Boring infrastructure for agent builders. 22 pay-per-call utility tools served as an MCP server — scrape, validate, embed, store, moderate, notify, convert, prevent loops. No accounts, no API keys, no subscriptions. Pay with USDC on Base via the x402 protocol.

npx @melis-ai/x402-tools-mcp

→ Full catalogue with prices and live examples: agents.melis.ai → Canonical 5-step RAG pipeline: agents.melis.ai/pipelines/rag → 5-minute getting started: agents.melis.ai/docs/getting-started

Tool catalogue (22)

Tool	Price (USDC)	Description
`list_services`	free	Returns the full catalogue with prices and endpoint URLs
`cacheserve`	$0.001	Fetch a URL with server-side caching — avoids redundant origin hits
`docconvert_text`	$0.001	Format conversion: md↔html, json↔csv, html→txt, etc.
`intentflow`	$0.001	Context handoff relay for multi-agent delegation (async, returns retrieve_url)
`memoryserve_write`	$0.001	Store text + embedding in Qdrant + SQLite, namespaced per `agent_id`
`memoryserve_query`	$0.001	Recall most-similar memories by semantic query
`memscrub`	$0.001	Scan retrieved RAG content for indirect prompt injection (10 patterns)
`notifyrelay_webhook`	$0.001	POST a JSON payload to a URL with optional HMAC-SHA256 signing
`schemagate`	$0.001	Validate a JSON response against a JSON Schema, return a hint on failure
`loopwall_hop`	$0.0005	Validate one hop of a signed multi-agent chain (loop / budget enforcement)
`loopwall_issue`	$0.001	Issue a signed Job Envelope at chain start with budget + hop cap
`imageguard`	$0.002	NSFW image classification (multi-class moderation on roadmap)
`notifyrelay_telegram`	$0.002	Send a Telegram message to a known chat ID
`promptguard`	$0.002	Score untrusted input for prompt injection risk (0–100)
`structextract`	$0.002	Extract structured JSON from HTML — tables, links, emails, phones
`xaudit`	$0.002	Validate API response content with a cryptographically signed certificate
`docconvert_pdf`	$0.005	Convert HTML or markdown to PDF (base64 output)
`kyaoracle`	$0.005	On-chain trust score (0–100) for an Ethereum/Base wallet address
`linkrisk`	$0.005	URL risk profile — heuristic phishing signals, redirect tracing
`markdownopt`	$0.005	Convert URL or HTML to clean LLM-ready markdown (~70% token reduction)
`notifyrelay_email`	$0.005	Send a transactional email via Resend (allowlisted, rate-limited)
`embedpay`	$0.00005 / 1k tokens	OpenAI text-embedding-3-small wrapper, batch tier available
`linksafe`	$0.01	Definitive URL safety check via Playwright sandbox + VirusTotal
`scrapepay`	$0.01	Web extraction via Playwright (robots.txt enforced, SSRF-safe)
`pdf_render`	$0.49	High-fidelity URL or HTML → PDF via Playwright

22 services. Live prices and live wallet on Basescan.

Why x402, why pay-per-call

No signup tax. Every tool here works the moment your wallet has USDC. No accounts to provision, no API keys to rotate, no subscription tiers to forecast.
Charge-on-success-only. Every service settles payment only on HTTP 2xx with non-empty content. Failed calls cost nothing — your retry budget is preserved.
Composes naturally. ScrapePay → MarkdownOpt → EmbedPay → MemoryServe → MEMSCRUB is the canonical RAG pipeline. Each step's payment is independent and verifiable on-chain.
MCP-native. One install, all 22 tools appear in Claude Desktop, Cursor, Cline, Continue, Claude Code, or any MCP-aware agent runtime.
No vendor lock-in. x402 is a Linux Foundation protocol (April 2026) co-signed by Google, AWS, Stripe, Visa, Mastercard. Other providers will offer x402 services; your wallet client works with all of them.

Installation — Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "x402-tools": {
      "command": "npx",
      "args": ["@melis-ai/x402-tools-mcp"]
    }
  }
}

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json Windows: %APPDATA%\Claude\claude_desktop_config.json

Restart Claude Desktop. All 22 tools appear in the tool picker.

Installation — Cursor, Cline, Continue, Claude Code

See agents.melis.ai/docs/install-mcp for per-client setup.

Example: canonical RAG pipeline

// All 22 tools registered as MCP tools — your agent calls them like any function
const page    = await callTool('scrapepay',   { url: 'https://example.com/article' });
const md      = await callTool('markdownopt', { html: page.content });
const vec     = await callTool('embedpay',    { input: md.markdown });
const stored  = await callTool('memoryserve_write', {
  agent_id: 'agent-1', content: md.markdown, vector: vec.embedding,
});

// Later — recall and guard against indirect injection
const hits    = await callTool('memoryserve_query', {
  agent_id: 'agent-1', query: 'what did we learn about X?',
});
for (const chunk of hits.matches) {
  const scrub = await callTool('memscrub', { content: chunk.content });
  if (scrub.risk_level === 'safe') useChunk(chunk);
}

Total cost per 5k-token page: ~$0.017 USDC. Detailed walkthrough: agents.melis.ai/pipelines/rag.

Funding the wallet

x402 settles in USDC on Base (Coinbase L2). Two paths:

Coinbase Wallet — buy USDC, switch to Base network, send to your agent's wallet.
Bridge from Ethereum mainnet — bridge.base.org.

$1.00 of USDC covers 100 ScrapePay calls, 200 PromptGuard calls, or one full RAG-pipeline ingestion run.

Discovery surfaces (for agent builders + AI consumers)

URL	Purpose
agents.melis.ai/llms.txt	Concise LLM-readable catalogue
agents.melis.ai/llms-full.txt	Full schemas + composition recipes
agents.melis.ai/openapi.json	OpenAPI 3.1 spec for all 22 endpoints
agents.melis.ai/ai-plugin.json	ChatGPT-plugin manifest
`https://<service>.melis.ai/.well-known/mcp.json`	Per-service MCP discovery
agents.melis.ai/agents-faq	Agent-builder FAQ (distinct from human FAQ)

Comparisons

Source & licensing

MCP wrapper source (this repo): MIT licensed.
Service code (the 22 services themselves): closed source for security.
Site content (agents.melis.ai docs): CC-BY-4.0 — attribute "melis x402 Tools" with a link to agents.melis.ai.