MCP Server Architecture
Model Context Protocol is the bridge between AIOHM's AI layer and the tools that can act in the outside world.
What MCP Does
MCP lets agents discover structured capabilities instead of relying on vague prompt-only instructions.
- Servers expose named tools
- Agents call those tools through a controlled interface
- Tenant-aware configuration decides which servers are available in each workspace
Typical MCP Responsibilities
- WordPress publishing and content management
- Mautic campaign and contact operations
- SEO inspection and optimization tasks
- Internal platform knowledge access
- Developer and platform introspection for controlled system tasks
Server Types
Laravel-native servers
These live inside the main application and are registered directly from the Laravel side.
External Node.js servers
These run as separate processes and expose tool interfaces back into the platform.
Why It Matters
MCP is what turns an assistant into an operator.
- Muse can trigger reliable tool calls
- Specialist agents can work with external systems in a structured way
- The platform can keep permission boundaries around what each tenant or agent is allowed to do
Operational Notes
- Server registration belongs to configuration and database state, not to the public docs build
- Production releases should ship the code needed to talk to MCP, but not local-only dev dependencies
- New MCP integrations should be documented here so operators know what capabilities are actually exposed