Architecture Overview
How CastHub MCP Works
From protocol to practice: how AI assistants control physical screens through standard protocols.
The MCP Protocol
CastHub MCP is built on the Model Context Protocol (MCP), an open standard created by Anthropic that lets AI models interact with external tools and data sources in a structured, predictable way. Instead of free-form API calls, MCP provides a formal contract between AI assistants and the systems they control.
How MCP Differs from REST APIs
Structured Tool Definitions
Each tool is defined with a schema that describes its purpose, inputs, and outputs. The AI understands what a tool does before invoking it, rather than parsing raw HTTP responses.
Typed Parameters
Every parameter has an explicit type, validation rules, and description. The AI knows that a screen ID is a string, a brightness level is an integer from 0 to 100, and a schedule requires a specific date format.
Semantic Context
MCP tools carry semantic meaning. The AI understands the relationship between tools, can chain operations logically, and provides human-readable previews before executing actions.
Why This Matters for AI-Native Integration
Traditional integrations require developers to build custom connectors for every AI platform. With MCP, CastHub exposes its entire signage control surface once, and any MCP-compatible assistant can immediately understand and use it. When Anthropic, OpenAI, or Microsoft adopt MCP support, they gain access to CastHub tools without any additional integration work.
This is the difference between building point-to-point integrations and building on a standard. Model Context Protocol is that standard for AI tool use.
CastHub's Implementation
CastHub's MCP server is built on .NET and hosted on Microsoft Azure. It exposes 58 discrete tools as MCP operations across ten domains: presentations, slide media, devices, device groups, schedules, alerts, device commands, dashboard, team members, and spaces. The team member and space tools need the Enterprise plan.
Authentication uses OAuth 2.0 with PKCE and dynamic client registration. Users sign in with their CastHub credentials, Google, or Microsoft, and authorize the AI client to act on their behalf. The server publishes RFC 9728 protected-resource metadata and supports RFC 7591 dynamic client registration so any compliant MCP client can connect without manual setup.
The same MCP tool surface that powers Claude also drives the ChatGPT and Microsoft Copilot connectors, all from a single source of truth. When a tool is added to the production server, it becomes available to every connected assistant without any manual specification authoring.
System Architecture
Dual API Backend
CastHub unifies two distinct backends behind a single AI-accessible interface. The CastHub Core APIs handle content management, presentation scheduling, screen grouping, and playlist operations. The Amazon Signage Remote Management API handles hardware-level operations like device restarts, screen captures, health monitoring, and display power control.
From the AI's perspective, these are all just tools. The MCP server abstracts the underlying API routing, so when an AI assistant asks to push content and then verify it rendered correctly via screen capture, both operations flow through the same tool interface even though they hit different backends.
Hardware Control via Amazon Signage API Live
CastHub bridges two APIs to give AI assistants unified control over both content and hardware. The Amazon Signage Remote Management API provides direct hardware control over Amazon Signage Stick devices, managed through the Amazon Signage platform.
Through CastHub's MCP server, AI assistants gain the ability to monitor device health, restart unresponsive hardware, capture what screens are currently displaying, and manage display power states, all without any human physically touching the device.
| Capability | What It Does | What It Means for AI Control |
|---|---|---|
| Factory Reset / Restart | Remotely restart or factory reset any Signage Stick | AI troubleshoots device issues without anyone touching hardware |
| Device Health Monitoring | CPU/GPU temps, Wi-Fi signal, memory, storage | AI detects problems before screens go dark |
| Screen Capture | Capture what's currently displaying | AI verifies content is rendering correctly |
| Device Information | MAC, serial number, IP address | AI has full device inventory awareness |
| Clear Application Cache | Clear cached content on the device | AI resolves stale content remotely |
| Display Control (HDMI-CEC) | Turn display on/off via HDMI-CEC | AI manages power schedules across all screens |
Note: Hardware control capabilities require Amazon Signage Stick devices. Content management and scheduling tools work with any CastHub-connected screen.
The Governance Layer
Every write operation in CastHub MCP follows a preview-confirm pattern. The AI client never executes a change blindly. Tools are annotated in the MCP server with ReadOnly, Destructive, or Idempotent hints. AI clients use those hints to ask for your confirmation before running anything that affects screens, devices, or content.
Preview-Confirm in Practice Live
When you ask an AI assistant to push a presentation to a group of screens, the flow looks like this:
Here's what your lobby screens will show:
This will display on Lobby-Main, Lobby-Left, and Lobby-Right. Confirm?
Risk Levels Live
Operations are categorized into four risk levels, each with its own confirmation requirements:
| Level | Examples | Confirmation |
|---|---|---|
| Level 1 (Read) | Screen status, list presentations, view schedules, screen capture, device health | Immediate |
| Level 2 (Standard Write) | Push content, update schedule, restart device, display on/off for group, clear cache | Preview + Confirm |
| Level 3 (Bulk Write) | Push to all screens, display on/off for all devices | Elevated Confirmation |
| Level 4 (Emergency / Destructive) | Emergency alert to all screens, factory reset devices | Dedicated Confirmation |
For the per-tool risk classification, see the tool reference.
Multi-Client Access
CastHub MCP tools are accessible from every major AI assistant through the same OAuth flow. The MCP server speaks one open protocol, so the same operations are available regardless of which client you use.
Cursor Live
Custom MCP server with OAuth, same URL works
https://webapi.cast-hub.com/mcp
Any MCP-compatible client Live
Custom MCP server with OAuth, RFC 9728 and RFC 7591 supported
Point your client at https://webapi.cast-hub.com/mcp
CastHub Chat Live
In-app chat widget built into the dashboard
Available in the CastHub dashboard
The MCP tool surface is shared across every connected client. When a new tool is added to the production server, it becomes available to every connected assistant without any manual specification authoring.
Authentication and Security
CastHub MCP uses OAuth 2.0 with PKCE and dynamic client registration for every supported AI client. There is no API key to copy and no header-based credential to manage. Users sign in with their CastHub email and password, Google, or Microsoft, and authorize the AI client to act on their behalf.
Key Security Principles
HTTPS Only
All communication between AI clients and the MCP server happens over HTTPS. There are no plaintext endpoints.
No Content Through LLM Providers
Your signage content, screen data, and device information travel directly between your AI client and CastHub's servers. Content never passes through or is stored by LLM providers (Anthropic, OpenAI, Microsoft).
Tool Risk-Level Annotations
Every tool is annotated as ReadOnly, Destructive, or Idempotent. AI clients use those hints to decide when to ask for your confirmation. Read tools run freely. Write and destructive tools request explicit confirmation before executing.
Standards-Based OAuth
The server publishes RFC 9728 protected-resource metadata at /.well-known/oauth-protected-resource/mcp and supports RFC 7591 dynamic client registration. DNS-rebinding defense via Origin validation is enforced.
Data Flow
The data flow is straightforward: your AI client sends a tool call to the CastHub MCP server. The server validates the OAuth token, applies the tool's risk-level annotation (preview-confirm for writes), executes the operation against the appropriate backend API, and returns the result directly to your AI client.
At no point does your screen content, device data, or organizational information leave the direct path between your client and CastHub's infrastructure. The LLM provider sees the tool call parameters and results, but CastHub's actual content payload is processed server-side.
Already live for every major AI assistant
Claude, ChatGPT, Microsoft Copilot, and Cursor all connect through the same OAuth flow today. Setup guide.
Ready to connect AI to your screens?
Get started with CastHub MCP and control your digital signage from any AI assistant.
Get Started