Connect Your Agent

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "acp": {
      "command": "python3",
      "args": ["path/to/mcp-server/server.py"],
      "env": {
        "ACP_API_KEY": "<your-api-key>",
        "ACP_BASE_URL": "https://controller.limitedview.co.uk/api"
      }
    }
  }
}

Add to .cursor/mcp.json in your project root:

// Cursor MCP setup (.cursor/mcp.json)
{
  "mcpServers": {
    "acp": {
      "command": "python3",
      "args": ["path/to/mcp-server/server.py"],
      "env": {
        "ACP_API_KEY": "<your-api-key>",
        "ACP_BASE_URL": "https://controller.limitedview.co.uk/api"
      }
    }
  }
}

Codex connects via stdio like Claude Code. Set the env vars and run:

# Codex / ChatGPT MCP setup
# 1. Install the MCP server
pip install mcp httpx pydantic

# 2. Set environment variables
export ACP_API_KEY="<your-api-key>"
export ACP_BASE_URL="https://controller.limitedview.co.uk/api"

# 3. Run the server (Codex connects via stdio)
python3 mcp-server/server.py

For ChatGPT Custom GPTs, import the OpenAPI spec as an Action:

# For ChatGPT Custom GPTs / Actions:
# Use the OpenAPI spec directly instead of MCP:
#
# OpenAPI spec URL:
#   https://controller.limitedview.co.uk/openapi/ai-control-plane-v1.yaml
#
# Authentication: API Key (Custom header)
#   Header name: X-Api-Key
#   Value: <your-api-key>
#
# Import the spec as a ChatGPT Action to enable
# all ACP endpoints as GPT capabilities.

services:
  acp-mcp:
    image: python:3.12-slim
    volumes:
      - ./mcp-server:/app
    working_dir: /app
    environment:
      - ACP_API_KEY=${ACP_API_KEY}
      - ACP_BASE_URL=https://controller.limitedview.co.uk/api
    command: python3 server.py
    stdin_open: true

• Input validation on all parameters (path traversal protection)
• Output sanitization (API keys, tokens, and secrets automatically redacted)
• HMAC-SHA256 signing on audit traces and financial data
• Write confirmation required for all mutations
• Rate limiting per operation category
• Stdio transport only (no network surface)
• Credentials stored locally with restrictive file permissions (600)
• All downloadable artifacts are signed with companion .sig.json