About x402station-mcp
x402station-mcp connects MCP-speaking agents to x402station risk signals before they pay an endpoint. Its tools report endpoint warnings and payment-related evidence, without routing payments or taking custody.
Key features
- 01Check endpoint risk before payment
- 02Retrieve seven-day endpoint forensics
- 03List currently flagged endpoints
- 04Report uptime, latency, and warnings
Use cases
- 01Screening an x402 URL before payment
- 02Investigating endpoint reliability and pricing
- 03Filtering flagged endpoints during discovery
MCP support
It exposes x402station Preflight, Forensics, Catalog Decoys, and other listed capabilities as tools for MCP clients. Paid tool calls use x402 and require a wallet private key holding Base mainnet USDC.
Configuration
Local server config is the same for every client. Copy it into your client’s MCP config file.
npx -y x402station-mcpAGENT_PRIVATE_KEYSecret0x-prefixed 64-hex-char private key. Account must hold Base mainnet USDC. Required for any paid tool call.X402STATION_BASE_URLOptional override. Only https://x402station.io or http(s)://localhost are accepted to prevent payload hijacking; any other value makes the adapter refuse to start.Local servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.
Tools
10Saved tool list · updated 2w ago
| Parameter | Type | Required | Description |
|---|---|---|---|
| url | string | Yes | The full URL of the x402 endpoint the agent is about to pay. |
Details
From the project READMEx402station-mcp
MCP adapter for x402station.io, the independent risk-signal layer for x402 agentic commerce. Exposes Preflight by x402station.io plus Forensics, Catalog Decoys, Alternatives, Credits, Watch, and Whats New. Any agent speaking the Model Context Protocol gets endpoint evidence before signing PAYMENT-SIGNATURE — decoy, zombie, price-trap, never-paid, latency, signature/settlement checks — before paying.
x402station.io independently probes every endpoint listed on agentic.market every 10 minutes and merges probe history with CDP settlement data. Policy engines decide and enforce; x402station.io measures and reports. We do not route, take custody, or endorse.
Install
# Claude Code / Cursor / Windsurf / Continue — works anywhere with MCP
npm install -g x402station-mcp
# or use npx in the config, no global install needed:
Configure
The adapter charges real USDC per paid call through x402 itself. You need a wallet private key that holds Base mainnet USDC.
Claude Code
Add to ~/.claude/claude_desktop_config.json (or wherever your MCP servers live):
{
"mcpServers": {
"x402station": {
"command": "npx",
"args": ["-y", "x402station-mcp"],
"env": {
"AGENT_PRIVATE_KEY": "0xYOUR_PRIVATE_KEY_HERE"
}
}
}
}
Cursor / Windsurf / Continue
Same shape — every MCP host understands command / args / env. See your tool's MCP docs.
Environment
| Variable | Required | Default | Purpose |
|---|---|---|---|
AGENT_PRIVATE_KEY | yes for any paid tool call | — | 0x-prefixed 64-hex-char private key. Account must hold Base mainnet USDC. |
X402STATION_BASE_URL | no | https://x402station.io | Override for dev / testing. |
Tools
preflight(url) — $0.001 USDC
Ask whether it's safe to pay this x402 URL. Returns:
{
"ok": true,
"warnings": [],
"metadata": {
"service": "...",
"price_usdc": "0.01",
"uptime_1h_pct": 100,
"avg_latency_ms": 412
}
}
ok is true only when no critical warning fires. Warnings include unknown_endpoint, no_history, dead, zombie, decoy_price_extreme, never_paid_zombie, proxy_markup, wildcard_402, spa_fallback, suspicious_high_price, slow, and new_provider.
forensics(url) — $0.001 USDC
Deep 7-day report. Superset of preflight. Returns hourly uptime, latency p50/p90/p99, status-code distribution, concentration-group stats, decoy probability. Extra warnings: dead_7d, mostly_dead, slow_p99, price_outlier_high, high_concentration.
catalog_decoys() — $0.005 USDC
Full blacklist. Returns every active endpoint currently flagged critical, plus per-reason counts. Pull periodically and cache locally — cheaper than preflighting every URL.
Typical agent flow
agent wants data from some_endpoint.com
│
├─ preflight("https://some_endpoint.com/data") ← $0.001
│ ok: false, warnings: ["decoy_price_extreme"]
│
└─ skip; try the next candidate
For bulk discovery, do catalog_decoys() once per day and treat the result as a set-difference against any URLs you're about to hit.
Composes with
x402station checks whether the endpoint is safe (decoy / zombie / price-trap / never-paid) before an agent pays. It pairs naturally with a payload safety check — screening the payment metadata for PII before signing:
agent intent: pay some_endpoint.com
├─ x402station preflight(url) ← is the ENDPOINT safe?
├─ presidio-x402 screen_payment_metadata() ← is the PAYLOAD safe?
└─ pay()
presidio-hardened-x402-mcp screens payment metadata for PII before signing. The two servers are developed independently, on purpose — keeping the signals uncorrelated is the point. This is a composition pointer, not an endorsement; see their repo for the screening tool's behaviour and limits.
Links
- Service: https://x402station.io
- Manifest: https://x402station.io/.well-known/x402
- OpenAPI: https://x402station.io/api/openapi.json
- Dataset: https://huggingface.co/datasets/x402station/preflight-dataset-v0_1
- GitHub: https://github.com/sF1nX/x402station-mcp
- x402 spec: https://x402.org
Contact
- General + commercial: [email protected]
- Bug reports: https://github.com/sF1nX/x402station-mcp/issues
- Security disclosures (RFC 9116): https://x402station.io/.well-known/security.txt
License
MIT
Comments
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.