Skip to main content
The Dakota Platform MCP server exposes platform data (customers, accounts, transactions) to AI clients like Claude Desktop, Claude Code, and ChatGPT via the Model Context Protocol.
The server is read-only today. Future versions will support orchestration commands (creating transactions, managing accounts).

​
Hosted Environments

EnvironmentURL
Productionhttps://mcp.dakota.xyz/mcp
Sandboxhttps://mcp.sandbox.dakota.xyz/mcp
Demohttps://mcp.demo.dakota.xyz/mcp
Developmenthttps://mcp.dev.dakota.xyz/mcp
Localhttp://localhost:5734/mcp
All hosted endpoints are HTTPS-only and require Stytch OAuth authentication.

​
Configuring Your AI Agent

​
Claude Desktop

Edit your config at ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows): 
{
  "mcpServers": {
    "dakota-platform": {
      "url": "https://mcp.dakota.xyz/mcp"
    }
  }
}
Restart Claude Desktop. The Dakota tools should appear in the tool picker, and you’ll be prompted to authenticate on first use.

​
Claude Code

claude mcp add dakota-platform https://mcp.dakota.xyz/mcp
Verify with claude mcp list. To remove: claude mcp remove dakota-platform.

​
ChatGPT / Other MCP Clients

Any MCP-compatible client can connect by pointing it at the /mcp endpoint above. The server speaks Streamable HTTP (HTTP + SSE) on a single endpoint:
  • POST /mcp — client messages
  • GET /mcp — SSE stream from server
  • Required client header: Accept: application/json, text/event-stream
  • Session managed via Mcp-Session-Id header

​
Local Development

For engineers running the server locally: 
cd platform-mcp
cp .env.example .env   # set PLATFORM_API_URL and SESSION_SECRET
make build
make run
Health check: curl http://localhost:5734/health Required env vars: PLATFORM_API_URL, SESSION_SECRET (min 32 chars). Optional: PORT (default 5734), LOG_LEVEL (default info), ENVIRONMENT (default development).

​
Available Tools

​
Customers

ToolDescription
list_customersList customers with filtering by name, type, KYB status, and creation date
get_customerGet detailed customer information including KYB links and provider statuses

​
Orchestration Accounts

ToolDescription
list_onramp_accountsList onramp accounts (bank transfer → crypto)
get_onramp_accountGet onramp account details
list_offramp_accountsList offramp accounts (crypto → fiat)
get_offramp_accountGet offramp account details
list_swap_accountsList swap accounts (crypto → crypto)
get_swap_accountGet swap account details

​
Transactions

ToolDescription
list_auto_transactionsList auto transactions with filtering by account, status, type, date range
get_auto_transactionGet auto transaction details including receipt and blockchain info
list_one_off_transactionsList one-off offramp transactions (single-use crypto → fiat)
get_one_off_transactionGet one-off transaction details

​
Example Prompts

Once connected, you can ask your AI agent things like:
  • “List the 10 most recent customers that completed KYB this month.”
  • “Show me all failed auto transactions for customer cust_xxx in the past 7 days.”
  • “What’s the total volume across all onramp accounts for cust_xxx?”
  • “Get details for offramp account acct_xxx and summarize its recent activity.”
The agent will pick the right tool, call the MCP server, and synthesize a response from the structured data.

​
Security Notes

  • All connections require authentication via Stytch OAuth.
  • The server validates Origin headers to prevent DNS-rebinding attacks.
  • Production endpoints are HTTPS-only.
  • The server is read-only — no mutations are possible today.
  • Never paste session tokens, API keys, or customer PII into prompts.

​
Reference