@cryptoapis-io/mcp-aml
Official# @cryptoapis-io/mcp-aml
MCP server for [Crypto APIs](https://cryptoapis.io/) AML (Anti-Money Laundering) product. Verify addresses and screen transactions for fraud, sanctions, and other AML risk categories.
> **API Version:** Compatible with Crypto APIs version **2024-12-12**
## Features
- Verify whether an address is flagged or associated with fraud, sanctions, ransomware, exploits, darknet markets, or other AML risk categories
- Screen a transaction and its participants against AML data, returning risk score, risk band, and flagged addresses with roles/categories/severity
- Get an address's exposure to risky entities/categories via the AML exposure graph
- List the paths connecting an address to risky entities in the AML exposure graph, with cursor pagination
## Prerequisites
- Node.js 18+
- [Crypto APIs](https://cryptoapis.io/) account and API key ([sign up](https://app.cryptoapis.io/signup) | [get API key](https://app.cryptoapis.io/api-keys))
## Installation
```bash
npm install @cryptoapis-io/mcp-aml
```
Or install all Crypto APIs MCP servers: `npm install @cryptoapis-io/mcp`
## Usage
```bash
# Run with API key
npx @cryptoapis-io/mcp-aml --api-key YOUR_API_KEY
# Or use environment variable
export CRYPTOAPIS_API_KEY=YOUR_API_KEY
npx @cryptoapis-io/mcp-aml
# HTTP transport (listens on 127.0.0.1; see "Exposing the server beyond localhost")
npx @cryptoapis-io/mcp-aml --transport http --port 3000 --api-key YOUR_API_KEY
```
### Claude Desktop
Add to your Claude Desktop config (`~/Library/Application Support/Claude/claude_desktop_config.json` on macOS, `%APPDATA%\Claude\claude_desktop_config.json` on Windows):
```json
{
"mcpServers": {
"cryptoapis-aml": {
"command": "npx",
"args": ["-y", "@cryptoapis-io/mcp-aml"],
"env": {
"CRYPTOAPIS_API_KEY": "your_api_key_here"
}
}
}
}
```
### Cursor
Add to `.cursor/mcp.json` (project) or `~/.cursor/mcp.json` (global):
```json
{
"mcpServers": {
"cryptoapis-aml": {
"command": "npx",
"args": ["-y", "@cryptoapis-io/mcp-aml"],
"env": {
"CRYPTOAPIS_API_KEY": "your_api_key_here"
}
}
}
}
```
### MCP Inspector
```bash
npx @modelcontextprotocol/inspector npx @cryptoapis-io/mcp-aml --api-key YOUR_API_KEY
```
### n8n
1. Start the server in HTTP mode:
```bash
npx @cryptoapis-io/mcp-aml --transport http --port 3000 --api-key YOUR_API_KEY
```
2. In your n8n workflow, add an **AI Agent** node
3. Under **Tools**, add an **MCP Client Tool** and set the URL to `http://localhost:3000/mcp`
> **n8n in Docker:** `localhost` inside the container is not your machine. Start the server with `--host 0.0.0.0` and `MCP_AUTH_TOKEN` set (see [Exposing the server beyond localhost](#exposing-the-server-beyond-localhost)), use `http://host.docker.internal:3000/mcp` as the URL, and add an `Authorization: Bearer <token>` header to the MCP Client Tool credential.
> All servers default to port 3000. Use `--port` to assign different ports when running multiple servers.
## Available Tools
### `aml`
| Action | Description |
|--------|-------------|
| `verify-address` | Check if an address is flagged for fraud, sanctions, or other AML risk categories. Params: `address` |
| `screen-transaction` | Screen a transaction and its participants against AML data. Params: `blockchain`, `transactionHash` |
| `get-address-exposure` | Get an address's exposure to risky entities/categories in the AML exposure graph. Params: `blockchain`, `address` |
| `list-address-exposure-paths` | List the paths connecting an address to risky entities in the AML exposure graph. Params: `blockchain`, `address`; optional `rootAddress` (only return routes originating from this flagged root address), `limit` (1-50, default 10), `startingAfter` (cursor — pass back the response's `nextStartingAfter` to page) |
`screen-transaction` supports: ethereum, ethereum-classic, binance-smart-chain, bitcoin, bitcoin-cash, dash, dogecoin, litecoin, arbitrum, polygon, avalanche, optimism, base, zcash, xrp, solana, tezos, kaspa, tron.
`get-address-exposure` and `list-address-exposure-paths` support a narrower set of 13 chains (no exposure graph for arbitrum, avalanche, base, optimism, tezos, or kaspa): ethereum, binance-smart-chain, ethereum-classic, polygon, tron, bitcoin, bitcoin-cash, litecoin, dogecoin, dash, zcash, xrp, solana.
> **Note:** these are the highest per-call credit cost endpoints in the Crypto APIs suite — `verify-address` is 200,000 credits, `screen-transaction` is 500,000 credits per call, `get-address-exposure` is 200,000 credits per request, and `list-address-exposure-paths` is 100,000 credits per page.
## CLI Arguments
| Argument | Description | Default |
|----------|-------------|---------|
| `--api-key` | Crypto APIs API key | `CRYPTOAPIS_API_KEY` env var |
| `--transport` | Transport type: `stdio` or `http` | `stdio` |
| `--host` | HTTP host (use `0.0.0.0` to accept remote connections — requires an auth token with `--api-key`) | `127.0.0.1` |
| `--auth-token` | Bearer token callers must send (`Authorization: Bearer <token>`); prefer the `MCP_AUTH_TOKEN` env var | `MCP_AUTH_TOKEN` env var |
| `--allowed-hosts` | Comma-separated `Host` header allowlist for non-loopback binds | — |
| `--port` | HTTP port | `3000` |
| `--path` | HTTP path | `/mcp` |
| `--stateless` | Enable stateless HTTP mode | `false` |
### HTTP API Key Modes
When using HTTP transport, the server supports two API key modes:
- **With `--api-key`:** The key is used for all requests. `x-api-key` request headers are ignored.
- **Without `--api-key`:** Each request must include an `x-api-key` header with a valid Crypto APIs key. This enables hosting a public server where each user provides their own key.
```bash
# Per-request key mode (multi-tenant)
npx @cryptoapis-io/mcp-aml --transport http --port 3000
# Clients send x-api-key header with each request
```
### Exposing the server beyond localhost
HTTP mode listens on `127.0.0.1` by default, so only processes on the same machine can reach it.
To accept connections from other machines or containers, bind explicitly and protect the port:
```bash
# Startup-key mode: callers must present the token (the server refuses to start without one)
export MCP_AUTH_TOKEN=$(openssl rand -hex 32)
npx @cryptoapis-io/mcp-aml --transport http --host 0.0.0.0 --port 3000 --api-key YOUR_API_KEY \
--allowed-hosts mcp.internal.example
# Clients send: Authorization: Bearer $MCP_AUTH_TOKEN
# Per-request key mode: no startup key, every request must carry the caller's own x-api-key
npx @cryptoapis-io/mcp-aml --transport http --host 0.0.0.0 --port 3000
```
`--allowed-hosts` restricts the `Host` header (DNS rebinding protection) when not bound to loopback. Prefer `MCP_AUTH_TOKEN` over `--auth-token`: command-line arguments are visible in the process list.
> Stdio transport always requires an API key at startup.
## Important: API Key Required
> **Warning:** Making requests without a valid API key — or with an incorrect one — may result in your IP being banned from the Crypto APIs ecosystem. Always ensure a valid API key is configured before starting any server.
## Remote MCP Server
Crypto APIs provides an official remote MCP server with all tools available via HTTP Streamable transport at [https://ai.cryptoapis.io/mcp](https://ai.cryptoapis.io/mcp). Pass your API key via the `x-api-key` header — no installation required.
## License
MIT
TDQS
Scored across 2 tools
The two tools are clearly distinct: `aml` performs blockchain AML risk checks and `system_info` returns reference documentation without API calls. An agent can easily choose between them. The multiple AML actions are bundled under one tool, but that is an intra-tool concern rather than tool-selection ambiguity.
`system_info` uses a snake_case noun phrase while `aml` is a bare acronym, so the two names do not follow a single predictable pattern. Both remain readable and reasonably descriptive, but the convention is mixed.
Two tools is on the thin side for the stated AML purpose, especially since one is purely reference documentation. The functional surface is effectively a single `aml` tool with four actions, making the overall offering borderline minimal rather than well-scoped.
The AML surface covers address verification, transaction screening, address exposure, and exposure-path discovery, which is solid coverage of the core AML workflow. Minor gaps exist, such as batch screening or broader entity lookup, but agents can work around them.