mcp-tools

The mcp-tools skill is a specialized framework for software agents to perform high-precision technical research and context gathering. It is designed for developers who need to bridge the gap between official framework documentation, real-world implementation patterns, and persistent codebase memory. By orchestrating specific tools like context7 and gh_grep alongside plugin-backed systems such as Supermemory, this skill ensures that agents remain focused, avoid context bloat, and maintain high standards of factual accuracy in their implementation work.

• Performs targeted lookups for API signatures, framework documentation, and library-specific upgrade notes using the context7 protocol.

• Mines production-grade implementation patterns and examples through gh_grep to cross-check best practices against official documentation.

• Coordinates research efforts by treating MCP servers as primary research tools and plugin-backed services as durable memory providers.

• Employs a rigorous research workflow that prioritizes narrow, concrete queries over broad or speculative searches to maintain agent session efficiency.

• Translates findings directly into the existing project style and codebase architecture rather than performing blind copies of external data.

• Strictly manages research boundaries by avoiding unnecessary context injection and preventing the storage of sensitive secrets or credentials in memory tools.

• This skill is intended for senior developers and automated coding assistants working on complex technical tasks within an OpenCode-configured repository.

• Always start with the narrowest possible query to minimize token consumption and irrelevant noise.

• Use context7 for official sources and switch to gh_grep only when official documentation leaves gaps or requires real-world verification.

• Remember that plugin-backed tools like Supermemory are meant for persistent context injection and codebase-wide recall, not as a replacement for live MCP-based research.

• Ensure all findings are summarized concisely; avoid dumping entire documentation pages into the session to prevent context window saturation.

• Adhere to the provided safety guidelines: do not use this skill to store credentials, and always prioritize local codebase reading before reaching for external search tools.