MCP Server Overview

MCTL exposes a Model Context Protocol (MCP) server that lets AI assistants manage your infrastructure through natural language.

What is MCP?

MCP is an open protocol that connects AI models to external tools and data sources. Instead of copy-pasting commands, you describe what you want in natural language and your AI assistant calls the appropriate tools.

Server Details

Endpoint	https://api.mctl.ai/mcp
Transport	Streamable HTTP (POST/GET)
Auth	OAuth 2.0 PKCE or Bearer token
Tools	Discover the tools available to your account with MCP tools/list

Capabilities

With the MCTL MCP server you can:

• Manage tenants — create, list, inspect, and delete tenants
• Deploy services — deploy, scale, rollback, and retire services
• Monitor operations — track async operations and workflows
• Handle incidents — list, inspect, acknowledge, and resolve incidents
• Manage domains — add, verify, and remove custom domains
• Provision databases — create databases with auto-configured secrets
• Preview environments — create and manage preview deployments
• Resource management — check CPU and memory usage
• Platform skill registry — list, read, publish, deprecate, and enable/disable platform-wide skills per tenant

What You Get

Once connected, your AI assistant can:

1. Inspect platform state — query services, tenants, resource usage, logs, and workflow status in real time
2. Deploy and manage apps — onboard, deploy, rollback, and preview services; create workspaces; provision databases
3. Manage domains and scaling — add custom domains with auto-TLS, configure autoscaling, verify DNS
4. Operation tracking — workflow-backed changes return a workflow name; other writes update their own registries or stores

How Access Works

Authentication follows a three-step model:

1. Token validated per request — the access token your client received when you signed in is checked on every MCP call. OAuth refresh-token hashes are stored server-side; the client manages its session.
2. Team-scoped access — tenant membership and roles are checked against the configured authorization authority. See Authorization.
3. Operation-specific effects — deployments use workflows and GitOps; domain registration, incident acknowledgements and other registry writes follow their own paths. Destructive changes may not be reversible.

For detailed auth information, see Authentication.

Tool Annotations

Every tool is annotated with its behavior:

Annotation	Meaning
readOnly	Only reads data, no side effects
write	Creates or modifies resources (deploys, scaling, config changes)
destructive	Potentially destructive effects; client confirmation behavior varies

See the full Tools Reference for details on each tool.