MCP Best Practices

Server Naming

• Python: {service}_mcp (e.g., slack_mcp)
• Node/TypeScript: {service}-mcp-server (e.g., slack-mcp-server)

Tool Naming

• Use snake_case with service prefix
• Format: {service}_{action}_{resource}
• Example: slack_send_message, github_create_issue

Response Formats

• Support both JSON and Markdown formats
• JSON for programmatic processing
• Markdown for human readability

Pagination

• Always respect limit parameter
• Return has_more, next_offset, total_count
• Default to 20-50 items

Transport

• Streamable HTTP: For remote servers, multi-client scenarios
• stdio: For local integrations, command-line tools
• Avoid SSE (deprecated in favor of streamable HTTP)

Server Naming Conventions

Follow these standardized naming patterns:

Python: Use format {service}_mcp (lowercase with underscores)

• Examples: slack_mcp, github_mcp, jira_mcp

Node/TypeScript: Use format {service}-mcp-server (lowercase with hyphens)

• Examples: slack-mcp-server, github-mcp-server, jira-mcp-server

The name should be general, descriptive of the service being integrated, easy to infer from the task description, and without version numbers.

Tool Naming and Design

Tool Naming

1. Use snake_case: search_users, create_project, get_channel_info
2. Include service prefix: Anticipate that your MCP server may be used alongside other MCP servers
   • Use slack_send_message instead of just send_message
   • Use github_create_issue instead of just create_issue
3. Be action-oriented: Start with verbs (get, list, search, create, etc.)
4. Be specific: Avoid generic names that could conflict with other servers

Tool Design

• Tool descriptions must narrowly and unambiguously describe functionality
• Descriptions must precisely match actual functionality
• Provide tool annotations (readOnlyHint, destructiveHint, idempotentHint, openWorldHint)
• Keep tool operations focused and atomic

Response Formats

All tools that return data should support multiple formats:

JSON Format (response_format="json")

• Machine-readable structured data
• Include all available fields and metadata
• Consistent field names and types
• Use for programmatic processing

Markdown Format (response_format="markdown", typically default)

• Human-readable formatted text
• Use headers, lists, and formatting for clarity
• Convert timestamps to human-readable format
• Show display names with IDs in parentheses
• Omit verbose metadata

Pagination

For tools that list resources:

• Always respect the limit parameter
• Implement pagination: Use offset or cursor-based pagination
• Return pagination metadata: Include has_more, next_offset/next_cursor, total_count
• Never load all results into memory: Especially important for large datasets
• Default to reasonable limits: 20-50 items is typical

Example pagination response:

Transport Options

Streamable HTTP

Best for: Remote servers, web services, multi-client scenarios

Characteristics:

• Bidirectional communication over HTTP
• Supports multiple simultaneous clients
• Can be deployed as a web service
• Enables server-to-client notifications

Use when:

• Serving multiple clients simultaneously
• Deploying as a cloud service
• Integration with web applications

stdio

Best for: Local integrations, command-line tools

Characteristics:

• Standard input/output stream communication
• Simple setup, no network configuration needed
• Runs as a subprocess of the client

Use when:

• Building tools for local development environments
• Integrating with desktop applications
• Single-user, single-session scenarios

Note: stdio servers should NOT log to stdout (use stderr for logging)

Transport Selection

Security Best Practices

Authentication and Authorization

OAuth 2.1:

• Use secure OAuth 2.1 with certificates from recognized authorities
• Validate access tokens before processing requests
• Only accept tokens specifically intended for your server

API Keys:

• Store API keys in environment variables, never in code
• Validate keys on server startup
• Provide clear error messages when authentication fails

Input Validation

• Sanitize file paths to prevent directory traversal
• Validate URLs and external identifiers
• Check parameter sizes and ranges
• Prevent command injection in system calls
• Use schema validation (Pydantic/Zod) for all inputs

Error Handling

• Don’t expose internal errors to clients
• Log security-relevant errors server-side
• Provide helpful but not revealing error messages
• Clean up resources after errors

DNS Rebinding Protection

For streamable HTTP servers running locally:

• Enable DNS rebinding protection
• Validate the Origin header on all incoming connections
• Bind to 127.0.0.1 rather than 0.0.0.0

Tool Annotations

Provide annotations to help clients understand tool behavior:

Important: Annotations are hints, not security guarantees. Clients should not make security-critical decisions based solely on annotations.

Error Handling

• Use standard JSON-RPC error codes
• Report tool errors within result objects (not protocol-level errors)
• Provide helpful, specific error messages with suggested next steps
• Don’t expose internal implementation details
• Clean up resources properly on errors

Example error handling:

Testing Requirements

Comprehensive testing should cover:

• Functional testing: Verify correct execution with valid/invalid inputs
• Integration testing: Test interaction with external systems
• Security testing: Validate auth, input sanitization, rate limiting
• Performance testing: Check behavior under load, timeouts
• Error handling: Ensure proper error reporting and cleanup

Documentation Requirements

• Provide clear documentation of all tools and capabilities
• Include working examples (at least 3 per major feature)
• Document security considerations
• Specify required permissions and access levels
• Document rate limits and performance characteristics

Source

• reference/mcp_best_practices.md(opens in a new tab)
• Raw file(opens in a new tab)
• anthropics/skills(opens in a new tab)