MCP Integration
Overview
Durantic includes a Model Context Protocol (MCP) server that enables AI assistants — such as Claude, ChatGPT, and other MCP-compatible clients — to manage your infrastructure through natural language. Instead of clicking through the dashboard or writing scripts, you can describe what you want and the AI carries it out using the Durantic API.
The MCP server exposes the same capabilities as the dashboard and REST API, letting AI assistants list machines, update configurations, trigger provisioning, execute commands, and more.
Setting Up
Claude Code (CLI)
Add the Durantic MCP server with a single command:
Claude Desktop
- Click the MCP Server button (plug icon) in the dashboard navigation bar.
- Copy the JSON configuration shown in the dialog.
- Paste it into your Claude Desktop MCP settings (
Settings > Developer > Edit Config).
Other MCP Clients
Add the Durantic SSE endpoint to your client's MCP server list:
The MCP server uses OAuth 2.1 — your client will open a browser window for authentication on first connection.
What Can the AI Do?
Once connected, the AI assistant can perform the following operations through the MCP server:
| Category | Capabilities |
|---|---|
| Machines | List machines, inspect machine details, view hardware and network information |
| Configuration | Update machine settings, assign and remove roles, manage variables and secrets |
| Provisioning | Trigger provisioning, monitor provisioning progress |
| Commands | Execute commands on machines and retrieve output |
| Roles | Create, edit, and delete roles; test role templates against machines |
| Mesh networking | Check mesh connectivity, view peer status, inspect endpoints |
| Blueprints | Discover and execute infrastructure blueprints |
Blueprints
Blueprints are pre-built infrastructure recipes that an AI assistant can execute on your behalf. They package multi-step workflows into a single, parameterized action.
Blueprint Sources
| Source | Description |
|---|---|
| Official | Maintained by Durantic. Cover common infrastructure patterns such as setting up a web server cluster or configuring a database replica set. |
| Community | Shared by other Durantic users. Available in a public catalog. |
| Private | Created by you or your team. Visible only within your account. |
Using a Blueprint
To use a blueprint, ask the AI assistant to run it and provide the required parameters. For example:
"Use the webserver-cluster blueprint to set up three machines with Nginx and a shared mesh network."
Behind the scenes, the AI calls use_blueprint with the blueprint name and parameters, and the MCP server orchestrates the multi-step workflow — creating roles, assigning them to machines, configuring mesh membership, and triggering provisioning as needed.
Security
MCP integration is secured through multiple layers:
| Layer | Description |
|---|---|
| OAuth 2.1 + PKCE | Authentication uses the OAuth 2.1 protocol with PKCE (Proof Key for Code Exchange) to prevent token interception. |
| Token scopes | Access tokens are scoped to specific operations, limiting what the AI can do. |
| Account-scoped operations | All actions are restricted to your account. The AI cannot access other accounts or resources outside your organization. |
| Audit trail | Every action performed through the MCP server is logged with a full audit trail, including the AI client identity, timestamp, and operation details. |
You can review MCP activity in your account's audit log to verify what actions the AI has performed.