MCP Integration

The Pulse MCP server exposes our market data and backtesting tools as LLM-callable functions via the Model Context Protocol. Ask your AI assistant to analyse spread behaviour, run a TWAP backtest, or profile order book depth — all in natural language.

Available tools

The following tools are registered on the MCP server. Your AI assistant will automatically discover them when connected.

Available data

See Download data for the full symbol list, available cached simulations, and output file details.

Discovering symbols and dates

Two tools answer "what can I run?", one per engine. Both catalogs are too large to hand an assistant in one response, so both summarise by default. Each instrument comes back as one row with date_count, first_date and last_date instead of the per-date detail. A symbol or date filter, or include_dates, returns the dates themselves.

list_calibrated_data lists what the agent-based model can simulate. Every instrument and date it returns is calibrated, so it is a valid input to run_simulation with the default engine.

It returns total (matches before paging), returned, limit, offset, truncated and instruments, so the assistant can tell when a listing was cut short and page for the rest.

list_available_data lists the raw data registry: every symbol-day that exists, calibrated or not. That is what a foundation model can be prompted with, and what could be calibrated.

It returns total, limit, offset and symbols, one entry per {symbol, provider, exchange}.

You rarely need to name these arguments yourself. Asking "which dates can I simulate 700.HK on?" or "what is ready on 2 September 2025?" is enough for the assistant to pick the right tool and filter.

Prerequisites

• A Pulse API key — get one from your Dashboard
• Node.js installed (for the mcp-remote bridge used by Claude Desktop)

Claude Code (CLI)

The quickest way to connect — add the server in one command:

claude mcp add pulse -s user \
  --transport http \
  https://pulse-mcp.simudyne.com/mcp \
  --header "X-API-Key: YOUR_API_KEY"

Restart Claude Code and you are ready to go. Ask it to "run a flash crash simulation for 700.HK".

Claude Desktop

1. Open your config file

On macOS it is located at:

~/Library/Application Support/Claude/claude_desktop_config.json

On Windows:

%APPDATA%\\Claude\\claude_desktop_config.json

2. Add the Pulse MCP server

Paste the following into the mcpServers object, replacing YOUR_API_KEY with your actual key:

{
  "mcpServers": {
    "pulse": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://pulse-mcp.simudyne.com/mcp",
        "--header",
        "X-API-Key: YOUR_API_KEY"
      ],
      "timeout": 300000
    }
  }
}

3. Restart Claude Desktop

Fully quit Claude Desktop (Cmd+Q on macOS) and reopen it. You should see a tools icon at the bottom of the chat input showing pulse with 20 tools available.

Try asking: "What symbols are available for simulation?"

Any MCP-compatible client

The Pulse MCP server works with any tool that supports the Model Context Protocol — including Cursor, Windsurf, VS Code + GitHub Copilot, Cline, Zed, and many others. To connect, you need three things:

What you need

• MCP server URL — https://pulse-mcp.simudyne.com/mcp
• Your API key — passed as an X-API-Key header with every request
• Node.js — required for the mcp-remote bridge that converts between stdio and HTTP (most desktop clients use stdio under the hood)

Generic configuration

MCP is a client- and model-agnostic protocol, so the same server works regardless of which assistant you use — only the config file location differs between clients. Replace YOUR_API_KEY with your actual key and paste this block into your client's MCP server configuration:

{
  "mcpServers": {
    "pulse": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://pulse-mcp.simudyne.com/mcp",
        "--header",
        "X-API-Key: YOUR_API_KEY"
      ],
      "timeout": 300000
    }
  }
}