MCP
1. Overview
The MCP Server screen connects AI assistants (Claude, GitHub Copilot, etc.) directly to your Nukk platform account through the Model Context Protocol (MCP). Instead of describing to the AI what exists in your environment, it starts actually querying and operating the platform, in real time, through the tools exposed by the MCP server.
In practice, this turns the AI assistant into an operator with direct access to your account — able to read the platform's current state and execute actions, always restricted to the scope of your own account.
2. Connection
The screen shows the data needed to connect an MCP client to your account:
| Field | Description |
|---|---|
| Status | Enabled / OAuth ready — indicates the MCP server is active and ready for OAuth authentication. |
| Endpoint | Your account's MCP server URL (e.g. https://br-1.nukk.com.br/account/nukk/mcp). |
| Client ID | OAuth client identifier used in the connection. |
| Client secret | OAuth client secret. Hidden by default (Reveal to show it, Rotate to generate a new one). |
| Issuer | OAuth authentication issuer URL (e.g. https://auth.br-1.nukk.com.br). |
The Refresh button updates the connection data shown on screen.
3. How to Connect
Add the Endpoint as a custom connector in your MCP client — for example, in Claude, Claude Code, or GitHub Copilot — and authenticate with your Nukk account. Authentication uses OAuth (Authorization Code + PKCE), so you don't need to manually copy/paste sensitive credentials on every connection: you authorize once, through the standard login flow.
Once connected, your AI assistant starts "seeing" your Nukk account through the tools made available by MCP, covering secrets, config maps, artifacts, deploys, routes, consumers, and observability — always restricted to your own account. No delete operation is exposed, so the assistant can query and act, but can't delete resources on the platform.
4. What You Can Do with MCP Connected
With MCP enabled, an assistant like Claude Code stops being just a pair-programmer over local code and starts operating the platform alongside you:
- Code your application from the assistant itself — ask it to write, review, or adjust the code of a Flow, Backend App, or Frontend App, already with the real context of what's published on the platform.
- Deploy a Flow or deploy an app right from the conversation, without switching screens.
- Check Config Maps and Secrets bound to an environment to quickly understand how an application is configured, without having to navigate the platform manually.
- Inspect Artifacts, Routes, and Consumers to understand what's exposed, in which environment, and with what security configuration.
- Check observability (logs, metrics) to investigate a problem without leaving the assistant's workflow.
Why This Matters
- For developers: reduces the back-and-forth between the code editor and the platform — the assistant already knows what's deployed, which secrets/config maps exist, and can act on it directly, speeding up debugging and iteration.
- For project managers: lets you ask for a snapshot of the environment's current state (what's deployed, with which resources, which routes exposed) in natural language, without depending on someone else navigating the platform to answer.
- For entire squads: standardizes how AI interacts with Nukk's infrastructure — any MCP-compatible assistant gets the same level of context and capacity to act, within the security limits of the connected account.
5. FAQ
Is MCP access limited to my account, or can I see data from other accounts/squads? Access is always restricted to the scope of the connected account itself.
Can the AI assistant delete resources via MCP? No — no delete operation is exposed by MCP's tools.
Do I need to generate a new Client secret periodically? TODO: confirm whether there's a recommendation/policy for rotating the Client secret, beyond the manual Rotate option.