MCP for Claude, ChatGPT and Cursor
A prompt is not a limit. A model can misread a number, repeat itself, or be talked out of an instruction. Guard’s MCP server gives an agent tools that cannot break your limits, because every order still passes the same nine rules as any bot’s.
The tools
Section titled “The tools”| Tool (planned) | What it does |
|---|---|
account | equity, positions, stops, open orders |
limits | your rules and the risk state (active, halted for the day, stopped) |
preview_order | “what would you allow for this trade?”: the size and the rule that bound it, without sending |
place_order | entry with its stop, through the nine rules |
amend_stop | move a stop; only tighter is accepted |
close_position | reduce or close; never blocked |
kill | pull the kill switch |
There is no tool to raise a limit, loosen a stop, resume after a halt or release the kill switch. Those need you, at the machine.
Example. You ask the agent to “go long ETH with 10x, stop 1% below”. It calls preview_order and gets back: “resized: open positions may be worth at most 5x equity”. It can report that to you; it cannot change it.
Claude Desktop
Section titled “Claude Desktop”Edit claude_desktop_config.json (macOS: ~/Library/Application Support/Claude/claude_desktop_config.json; Windows: %APPDATA%\Claude\claude_desktop_config.json), then restart Claude Desktop.
{ "mcpServers": { "zunder-guard": { "command": "zunder-guard", "args": ["mcp"] } }}filled in from your settings ·
Claude Code
Section titled “Claude Code”claude mcp add zunder-guard -- zunder-guard mcpThe -- separates Claude Code’s options from the server’s command.
Cursor
Section titled “Cursor”.cursor/mcp.json in a project, or ~/.cursor/mcp.json for all projects:
{ "mcpServers": { "zunder-guard": { "command": "zunder-guard", "args": ["mcp"] } }}OpenAI Agents SDK
Section titled “OpenAI Agents SDK”from agents.mcp import MCPServerStdio
async with MCPServerStdio( name="zunder-guard", params={"command": "zunder-guard", "args": ["mcp"]},) as guard: ... # pass `guard` in the agent's mcp_serversChatGPT
Section titled “ChatGPT”ChatGPT’s connectors reach MCP servers over the internet (SSE or streamable HTTP); a local server started from a command does not work there. Guard runs on your machine and listens on localhost only. So ChatGPT is not supported directly.
We do not recommend exposing Guard to the internet with a tunnel to make it work. A connection through the relay is being considered; it is not planned yet.
What the MCP server needs
Section titled “What the MCP server needs”zunder-guard mcp talks to your running Guard on your machine. It holds no key of its own: the API wallet key stays in Guard.
Sources
Section titled “Sources”- Claude Desktop: modelcontextprotocol.io, “Connect to local MCP servers”.
- Claude Code: code.claude.com/docs/en/mcp.
- Cursor: cursor.com/docs/context/mcp.
- OpenAI Agents SDK: openai.github.io/openai-agents-python/mcp.
- ChatGPT: OpenAI, “ChatGPT Developer mode”.