Connect your AI assistant via the MCP server

Overview

PreStartGo ships a Model Context Protocol (MCP) server that lets any MCP-aware AI assistant — Claude Desktop, Cursor, Continue, Cline, and others — query your fleet read-only over a stdio transport.

The server exposes 16 tool categories: equipment, checks, defects, services, sites, groups, tags, devices, alerts, observations, compliance, checklists, zones, documents, reports, and telemetry. Tool calls hit the same REST API your dashboard uses, scoped to the organisation that owns the API key.

This is read-only by design. You cannot create or modify records via MCP. If you need write access for an integration, use a webhook subscription instead.

Generate an API key

1. Sign in to admin.prestartgo.com.
2. Open Developer → API Keys.
3. Click Generate Key. Tick only the Read scopes for what the assistant should see.
4. Copy the key — it starts with psg_live_ for production or psg_test_ for sandbox. You will only see it once.

Configure Claude Desktop

Open Claude Desktop's settings file:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%/Claude/claude_desktop_config.json

Add the PreStartGo server under mcpServers:

{
  "mcpServers": {
    "prestartgo": {
      "command": "npx",
      "args": ["-y", "@prestartgo/mcp-server@latest"],
      "env": {
        "PSG_API_KEY": "psg_live_..."
      }
    }
  }
}

Restart Claude Desktop. The PreStartGo tools appear under the slash-tools menu.

Configure Cursor

Cursor reads MCP servers from ~/.cursor/mcp.json:

{
  "mcpServers": {
    "prestartgo": {
      "command": "npx",
      "args": ["-y", "@prestartgo/mcp-server@latest"],
      "env": {
        "PSG_API_KEY": "psg_live_..."
      }
    }
  }
}

Reload Cursor. The tools appear automatically.

Try it

Once connected, open a chat and ask:

• "Show me yesterday's bypass events across the Karratha site."
• "Which excavators have an open critical defect right now?"
• "List services due in the next 7 days, grouped by mechanic."
• "How many pre-starts did the Brisbane fleet complete last month?"

The assistant will call the appropriate tools, format the results, and let you follow up with questions. You can also pipe the output into reports, charts, or ticket creation in your other tools.

Security model

• API keys are organisation-scoped. A key cannot read data from other tenants.
• Keys are rotatable from the dashboard. Old keys revoke immediately on rotation.
• All requests are logged in your immutable audit trail with the key id.
• The MCP server is read-only. Write operations require either the dashboard, the REST API with a write-scoped key, or a webhook subscription.
• The transport is stdio (local process) — no inbound network listener.
• Mandatory MFA on the Account Owner and Administrator roles still applies for key generation.

Troubleshooting

• "No tools appear": confirm npx is on your PATH and that the API key starts with psg_live_ or psg_test_. Check the assistant's MCP-server log for connection errors.
• "401 Unauthorized": the key has been revoked, has passed its expiry date, or was copied wrongly. Generate a new key with read scopes. (A key missing a scope gets 403 instead.)
• "Empty results": the key is scoped to a different organisation than the one your data lives in. Switch organisations in the dashboard before generating the key.

Related

• Choosing between MCP, webhooks and the REST API: Webhooks, the API, or MCP?
• Key lifecycle, scopes and rotation: Authentication and API Keys
• Audit logs of MCP tool calls live in the same place as dashboard activity — see Audit trails.