mcp-guide

The mcp-guide skill serves as a strategic decision-making engine for developers and users of the OpenClaw agent platform. Its primary purpose is to help determine whether a given task is best executed via an MCP (Model Context Protocol) server or a dedicated, direct-API-based skill. By providing a structured evaluation matrix, this skill prevents common pitfalls such as excessive token overhead, unnecessary cold start latencies, and unexpected costs associated with default model configurations in third-party MCP servers.

• Provides clear heuristics for choosing between direct API integration (Skills) and MCP servers based on service complexity, cost sensitivity, and frequency of use.

• Includes a comprehensive decision matrix comparing latency, tool discovery, maintenance, and predictability across various integration types.

• Offers specific guidance for known services like Perplexity, PostHog, GitHub, and Google Workspace, highlighting when to bypass generic MCP setups in favor of optimized, custom-coded skills.

• Identifies common MCP anti-patterns, including the risks of expensive model defaults and the impact of schema validation on small, high-frequency queries.

• Use this skill when configuring new agent capabilities to ensure long-term architectural health and cost efficiency.

• Consult the guide when experiencing high latency in tool execution or when model billing exceeds expectations due to automated tool chains.

• Designed for technical users working with the OpenClaw framework who need to manage dynamic workspace configurations, mcporter.json files, and core skill deployments.

• Always favor direct skills for high-frequency or cost-sensitive tasks, while reserving MCP servers for complex, multi-step workflows or services with numerous, dynamic endpoints that benefit from automated discovery.