Connect
MCP
Connect Claude, Codex or VS Code to Novie Control through the Model Context Protocol (MCP), so you can read and manage it from the AI tool you already use.
Novie runs an MCP gateway: a Model Context Protocol endpoint that lets AI tools such as Claude, Codex and VS Code work with Novie Control on your behalf. These docs call such a tool an MCP client. Ask Claude "Which workloads in Marketing are close to their budget?" or "Top up the onboarding workload by $200", and it calls Novie tools with your identity and permissions.
Endpoint
https://mcp.novamindlabs.ai/mcpThe gateway uses the Streamable HTTP transport and standard MCP OAuth. You only need the URL: a compatible MCP client discovers the authorisation server, registers itself, and opens a browser for you to sign in.
Connect your AI tool
In Claude Code and Codex, the easiest way is to search for Novie in the official plugin marketplace and install it from there. For other tools, or to set it up by hand, use the steps below.
Claude Code
claude mcp add --transport http novie https://mcp.novamindlabs.ai/mcpThen run /mcp inside Claude Code and choose Authenticate.
Claude (desktop and web)
Open Settings → Connectors → Add custom connector, name it Novie and paste the endpoint URL. Sign in when prompted.
Codex
codex mcp add novie --url https://mcp.novamindlabs.ai/mcp
codex mcp login novieVS Code (GitHub Copilot)
Add the server to your user mcp.json:
{
"servers": {
"novie": { "type": "http", "url": "https://mcp.novamindlabs.ai/mcp" }
}
}How access works
- Sign-in: OAuth authorisation code with PKCE. Your token identifies you, not an organisation, so one connection works across every organisation you belong to; each tool call names the organisation it acts on.
- Permissions are checked on every call: Novie checks your current membership and role each time, so granting or revoking access takes effect on the next request without reconnecting.
- Sessions: access tokens last up to 24 hours and refresh automatically. If your session expires or is revoked, sign in again from your AI tool.
- Tool names are stable: your AI tool can cache the tool list safely. If a tool's inputs change incompatibly, the gateway asks it to refresh.
Reads and confirmed writes
Read tools (organisations, structure, budgets, workloads, usage, policies and more) are available in every connected AI tool, filtered by your permissions.
Write tools, such as creating a department, renaming a workload or activating a key, always ask for explicit confirmation. Your AI tool shows a preview of exactly what will change, and nothing happens until you confirm. Each write is idempotent and audited with the tool that made it. Destructive operations (retiring a department, revoking a key, archiving a subscription) are marked as destructive and also require confirmation.
Your organisation can restrict which tools may be called with MCP policies in the Policy centre.
What's next: one MCP gateway for the enterprise
In the next release, the Novie MCP gateway becomes your company's single MCP gateway. Every downstream AI tool and agent (Claude, Codex, Copilot, your own agents) configures one MCP server, Novie, and through it discovers and uses the tools your policies allow:
- One configuration: no per-tool MCP setup on every laptop or agent. Add or remove an upstream tool once, centrally.
- Policy-governed discovery: each person or agent sees only the tools that policy permits for them, their unit and their workload. Different callers see different tool lists from the same endpoint.
- Policy-governed use: every call is checked against current permissions and policy, can require approval, and is audited.
- Credentials stay central: upstream credentials live inside Novie and are never handed to the AI tools that connect.
Until then, add any third-party MCP servers to each AI tool yourself, as you do today.