Delta Exchange MCP

What is MCP?

A plain-language explainer of the Model Context Protocol, how the Delta Exchange server fits the client → server → API architecture, and why it suits trading workflows.

The Model Context Protocol (MCP) is an open standard, introduced by Anthropic in late 2024, for connecting AI applications to external tools and data through one uniform interface. Think of it as a USB-C port for AI: instead of every app inventing its own integration, any MCP-compatible client can plug into any MCP server and immediately use what it offers. The protocol is documented at modelcontextprotocol.io.

The Delta Exchange MCP server is one such server. It exposes Delta's market and account data — and, when you opt in, trading — as MCP tools, so any compatible AI client can use it in plain language.

The three pieces

An MCP setup is always three parts:

  1. Client (host): the AI application you already use, such as Claude Desktop, Cursor, or Claude Code. It holds the model and decides when to call a tool.
  2. Server: a small program that advertises a set of tools and answers calls. Here that is delta-exchange-mcp.
  3. Upstream resource: whatever the server talks to. Here that is Delta Exchange's REST API.

The client and server speak MCP to each other; the server speaks Delta's normal HTTPS API to Delta. Your credentials are used only on that last hop, from your machine to Delta. They never reach the model.

Tools, and who decides to call them

A server advertises tools: named operations with typed parameters, like get_ticker(symbol) or get_positions(...). You almost never name a tool yourself. You ask a question in plain language ("what's BTC funding right now?"), the model picks the right tool and arguments, the server runs it, and the model reads the result back and answers you.

That is why the same Delta server works identically across every client: the tools are the contract, and each client wires them to its own model.

You stay in plain language

The tool names and parameters in the Tool Reference are there for when you build agents or want to know exactly what runs. For everyday use, you just ask.

How this server connects

MCP servers can run in two shapes: a local subprocess over stdio, or a hosted service over HTTP (often with OAuth). The Delta Exchange server is the local kind:

  • It runs as a subprocess on your own machine and talks to your client over stdio. It is not a hosted service.
  • Your API key and secret are read from your local environment at launch and used only to sign requests sent directly to Delta's API.
  • They never pass through the AI model or any Delta server beyond the normal trading API.

See Security for the full data-flow and key model.

Why MCP for trading

  • Ask, don't integrate. Get live market and account data in plain language, with no glue code, SDK wiring, or dashboards to build.
  • Write once, use everywhere. One server works across every MCP client, so a prompt that works in Claude Desktop works in Cursor or Claude Code too.
  • Read-only by default. Out of the box every tool is a read-only GET; trading registers only when you opt in with DELTA_MCP_MODE=trade and a Trading-permission key. The server never has a withdrawal path.
  • Keys stay local. Credentials live in your client config and are used only on your machine. Nothing sensitive flows to the model.
  • Composable. Drop the same tools into an agent, a notebook, or a scheduled job for monitoring, P&L, and reconciliation.

Supported clients

The server works with any MCP-compatible client, including Claude Desktop, Claude Code, Cursor, VS Code, Zed, and Windsurf. Per-client setup is in the Quickstart.

Not investment advice

Output is generated by an AI model and can be inaccurate, incomplete, or non-deterministic. It is not investment advice. Trading crypto derivatives involves substantial risk. Delta Exchange is registered with the Financial Intelligence Unit (FIU), Govt. of India.

On this page