> ## Documentation Index
> Fetch the complete documentation index at: https://docs.getsly.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# MCP server

> @sly_ai/mcp-server — Sly capabilities as MCP tools for Claude Desktop and compatible runtimes.

The Sly MCP server exposes the full Sly API as Model Context Protocol tools. Install it into any MCP-compatible client and the agent (Claude, GPT, custom LLMs) gets native tool access to accounts, agents, wallets, transfers, streams, and all six commerce protocols.

## Install for Claude Desktop

Simplest path:

```bash theme={null}
npx @sly_ai/mcp-server install
```

This edits your Claude Desktop config file at:

* **macOS:** `~/Library/Application Support/Claude/claude_desktop_config.json`
* **Windows:** `%APPDATA%\Claude\claude_desktop_config.json`
* **Linux:** `~/.config/Claude/claude_desktop_config.json`

Then restart Claude. Sly tools appear under the 🔨 menu.

## Manual install

Add to your MCP client's config:

```json theme={null}
{
  "mcpServers": {
    "sly": {
      "command": "npx",
      "args": ["@sly_ai/mcp-server", "run"],
      "env": {
        "SLY_API_KEY": "pk_test_...",
        "SLY_ENVIRONMENT": "sandbox"
      }
    }
  }
}
```

Environment variables:

* `SLY_API_KEY` (required) — your API key
* `SLY_ENVIRONMENT` (optional) — `sandbox` (default), `production`
* `SLY_AGENT_TOKEN` (alternative) — agent token instead of API key
* `SLY_LOG_LEVEL` (optional) — `debug`, `info`, `warn`, `error`

## Tools exposed

The server exposes every tool appropriate for the calling credentials. Full list on the [MCP tool catalog](/agents/mcp-tool-catalog) page.

Highlights:

* Accounts + agents CRUD
* Wallet management + balances
* Transfers + streams
* Full protocol coverage (UCP, ACP, AP2, x402, A2A, MCP, MPP)
* Agent wallet policies + evaluation
* Governance (approvals, kill-switch, rejection explanations)

## Remote HTTP MCP

If your runtime prefers HTTP to stdio, Sly hosts an MCP endpoint directly:

```bash theme={null}
# JSON-RPC over HTTP
curl -X POST https://api.getsly.ai/mcp \
  -H "Authorization: Bearer agent_..." \
  -d '{ "jsonrpc":"2.0", "method":"tools/list", "id":1 }'

# Server-Sent Events streaming
curl -N -H "Authorization: Bearer agent_..." https://api.getsly.ai/mcp
```

## Permission filtering

The server only exposes tools the calling credentials can actually invoke. An agent at KYA Tier 0 won't see Tier-2-only tools. This keeps the LLM's tool catalog focused and prevents wasted, guaranteed-to-fail tool calls.

## Scanner MCP (separate package)

The Scanner service has its own MCP server with demand-intelligence tools (merchant readiness, protocol adoption, shopping simulations):

```bash theme={null}
npx @sly_ai/scanner-mcp install
```

19 tools, including `scan_merchant`, `batch_scan`, `get_readiness_report`, `run_agent_shopping_test`.

## Building a custom MCP integration

If you're wrapping Sly inside your own agent framework, you don't need the stdio server — use the [Node SDK](/sdks/node-sdk) and integrate directly. The MCP server is a convenience for drop-in MCP clients; it doesn't add capability.
