crypto-link-check
# Crypto Link Check
**Is that MetaMask / Phantom / Ledger / exchange link real?** Check it before you (or your agent) connect a wallet or sign anything.
October 2026 alone brought fake Phantom, Jupiter and Bitget browser extensions, a hijacked NGRAVE push alert that sent users to a seed-phrase page, and exchange phishing mails. AI agents can now trade from exchange accounts and pay for things on their own, so they click links too. This tool gives them (and you) a quick, read-only second opinion.
- **Free, offline:** CLI and MCP server that compare a link with the official domains of ~60 wallets, exchanges and dapps and flag lookalike characters (`0`/`o`, `rn`/`m`, Cyrillic), typosquats, punycode, free hosting and bait words like "claim" or "sync".
- **Free API + remote MCP:** `https://crypto-link-check.nextill-pos.workers.dev`
- **Paid full check (optional, 0.005 USDC via [x402](https://x402.org) on Base):** adds domain registration age (RDAP), DNS, redirect tracing and one fetch of the page scanned for seed-phrase prompts and token-approval drainer code. 1-10 URLs per call.
It never asks for keys or seed phrases. Verdicts are heuristics, not guarantees.
## CLI
```bash
npx devx-link-check phantorn.app metamask.io
# verdict: dangerous phantorn.app
# [danger] phantorn.app spells the official phantom.app with lookalike characters ...
# verdict: official metamask.io
```
Exit code: `0` official / no red flags, `2` caution / suspicious / dangerous, `1` error. `--json` for machine output.
## MCP
Local (stdio), runs offline:
```json
{ "mcpServers": { "crypto-link-check": { "command": "npx", "args": ["-y", "devx-link-check", "mcp"] } } }
```
Remote (streamable HTTP, no key): `https://crypto-link-check.nextill-pos.workers.dev/mcp`
Tools: `check_crypto_link(url)` and `full_report_info()`.
## HTTP API
| Route | Price | What |
|---|---|---|
| `GET /api/basic?url=...` | free | lexical verdict + findings |
| `GET /v1/check?url=...&url=...` | 0.005 USDC (x402 v2 `exact`, `eip155:8453`) | full check, 1-10 URLs |
| `GET /brands.json` | free | the official-domain list we use |
| `GET /openapi.json`, `/.well-known/x402.json` | free | discovery |
The paid route answers `402` with a `PAYMENT-REQUIRED` header; any x402 client can pay. Settlement happens only after the report is produced. Humans can pay from a browser wallet on the [web page](https://crypto-link-check.nextill-pos.workers.dev). Terms: [/terms](https://crypto-link-check.nextill-pos.workers.dev/terms).
Verdicts: `official`, `no_red_flags`, `caution`, `suspicious`, `dangerous`. Same-name-different-TLD domains are only `caution` because the brand may own them.
## Official-domain list
`cli/src/brands.js`. It is curated by hand and incomplete. PRs that add brands or official domains (with a source) are welcome.
## Layout
- `cli/`: npm package `devx-link-check` (CLI + stdio MCP)
- `worker/`: the Cloudflare Worker behind the hosted API
MIT. Made by [DevX](https://madebydevx.com). Contact madebydevx@gmail.com.
## Guide
- [How to spot fake crypto links](https://crypto-link-check.nextill-pos.workers.dev/how-to-spot-fake-crypto-links)
- Related free tool: [7702 Sweeper Check](https://sweeper-check.nextill-pos.workers.dev/) — is a sweeper on your wallet?
- Related free tool: [Address Poisoning Check](https://address-poisoning-check.nextill-pos.workers.dev/) — is the address you are about to pay a poisoned lookalike?
TDQS
Scored across 2 tools
The two tools are mostly distinct: one performs a local link check and the other only returns metadata about a paid hosted service. However, the name 'full_report_info' could be mistaken for an actual deep-scan operation, and only the description prevents that misselection.
Both names use snake_case and are readable, but they follow different structural conventions: an action verb phrase (check_crypto_link) versus a noun-ish descriptor (full_report_info). Minor deviation rather than inconsistency.
Two tools is thin for a security-checking server; a single checking tool plus a metadata/info tool is borderline. The narrow scope keeps it defensible, but it feels under-scoped.
There is a notable dead end: the server advertises a paid hosted check but exposes no tool to actually invoke it, only an info endpoint. There is also no batch-check tool or way to list supported brands, so agents cannot complete the advertised deeper-verification workflow.