Skip to main content
The @sly_ai/mcp-server package exposes the full Sly API as MCP tools. Install it into Claude Desktop, Cursor, or any MCP-compatible client and the agent instantly has access to accounts, agents, wallets, transfers, streams, protocols, and more.

Install

Claude Desktop:
This edits your Claude Desktop config. Restart Claude; the Sly tools appear in the tool catalog. Other clients (manual):

Remote HTTP MCP

If your runtime prefers HTTP over stdio:
Or SSE streaming for long-running tools:

Tool categories

Tools are grouped by resource. All respect the calling agent’s KYA tier and wallet policy.

Accounts

Agents

Wallets

Agent wallets + policies

Transfers & settlement

Protocol tools

UCP: sly_ucp_create_checkout, sly_ucp_get_checkout, sly_ucp_update_checkout, sly_ucp_complete_checkout, sly_ucp_cancel_checkout, sly_ucp_list_checkouts, sly_ucp_batch_checkout, sly_ucp_batch_complete, sly_ucp_add_payment_instrument, sly_ucp_add_fulfillment_event, sly_ucp_cancel_order, sly_ucp_update_order_status, sly_ucp_get_order, sly_ucp_list_orders, sly_ucp_discover ACP: sly_acp_create_checkout, sly_acp_get_checkout, sly_acp_complete_checkout, sly_acp_list_checkouts, sly_acp_batch_checkout AP2: sly_ap2_create_mandate, sly_ap2_get_mandate, sly_ap2_list_mandates, sly_ap2_update_mandate, sly_ap2_execute_mandate, sly_ap2_cancel_mandate x402: sly_x402_create_endpoint, sly_x402_get_endpoint, sly_x402_list_endpoints, sly_x402_pay, sly_x402_verify, sly_agent_x402_sign A2A: sly_a2a_send_task, sly_a2a_get_task, sly_a2a_list_tasks, sly_a2a_discover_agent MPP: sly_mpp_open_session, sly_mpp_get_session, sly_mpp_list_sessions, sly_mpp_close_session, sly_mpp_pay, sly_mpp_verify_receipt, sly_mpp_list_transfers

Governance + compliance

Agent funding (sandbox helpers)

Sly does not subsidise agent EOA top-ups. Tenants fund their own agent wallets: on testnet via the free Circle W3S faucet (above), on mainnet by transferring USDC the tenant already owns directly to the agent’s EOA address on-chain.

Example invocation

Response (success):

Permission filtering

The MCP server only exposes tools the calling agent/credentials can actually invoke. An agent at KYA Tier 0 won’t see tools that require Tier 2 even if they’re installed — reducing confusion for the LLM and preventing wasted tool calls.

Scanner MCP (separate server)

The Scanner service has its own MCP server with 19 tools for merchant readiness intelligence:
Tools include scan_merchant, batch_scan, get_readiness_report, get_protocol_adoption, get_heat_map, find_best_prospects, compare_merchants, generate_snapshot, and run_agent_shopping_test.

For agent authors

If you’re building an agent that uses MCP tools: the server automatically translates the LLM’s natural-language tool calls into Sly API requests. You don’t write HTTP code — you write good prompts, and the model picks the right tool from the catalog. Good practice:
  1. Give the agent your intended purpose in the system prompt (e.g. “You help pay supplier invoices. Use the Sly tools to check balances and send transfers.”)
  2. The agent will autonomously pick among the tools
  3. Wallet policies enforce what’s actually allowed — so even a confused LLM can’t send a 10ktransferfromanagentwitha10k transfer from an agent with a 500/day cap