Commerce MCP Server
5.1 Tool Design for an E-Commerce MCP Server
A complete e-commerce MCP Server typically includes the following tools:5.2 Tool Design Principles
Single Responsibility
Each tool should do one thing. Do not combine search and ordering in the same tool. The AI model will automatically compose multiple tools based on context.Clear Descriptions
A tool’sdescription directly affects when the AI model chooses to use it. Be explicit about:
- What the tool does
- When it should be used
- What type of data it returns
Sensible Parameter Design
- Keep required parameters minimal (lower the barrier to use)
- Provide defaults for optional parameters
- Write specific parameter descriptions
Leverage structuredContent
For return data that needs programmatic processing (such as price calculations, inventory comparisons), provide bothcontent (for LLM reading) and structuredContent (for programmatic processing):
Use isError to Distinguish Business Errors
For business issues encountered during tool execution (e.g., product not found, insufficient inventory), useisError: true instead of throwing an exception:
5.3 Complete E-Commerce MCP Server Example
5.4 Commerce Resource Design
Beyond tools, an e-commerce Server can also expose static context as Resources, enabling AI agents to understand your business:5.5 Permissions and Security
Commerce Servers require special attention to access control:
For remote deployment, use OAuth 2.1 authentication to ensure correct user identity and permissions. For tools that require authentication, you can request user login via Elicitation, or rely on the Host application’s existing OAuth integration.
5.6 Commerce MCP Server and Its Relationship to Other Protocols
In a complete Agentic Commerce system, MCP serves as the “connection layer” between AI applications and your commerce systems, while UCP/ACP/OTR handle the specific commerce protocol logic.
Next Chapter: Resources and Prompts — Deep dive into Resources and Prompts