mcp-research

The mcp-research skill empowers agents to bypass stale training data by leveraging live, source-backed technical documentation and web search capabilities. It is designed for software engineers, developers, and technical researchers who need verified, real-time data to make informed implementation decisions. Whether you are working with rapidly evolving frameworks like FastAPI or React, managing complex dependency trees, or debugging version-specific behavior, this skill provides a structured methodology for gathering accurate information.

• Resolves library IDs and queries official documentation using Context7 for authoritative API signatures and usage patterns.

• Retrieves code-centric examples from GitHub, Stack Overflow, and technical docs via Exa for practical implementation insights.

• Performs broad web searches for ecosystem updates, breaking changes, changelogs, and recent community announcements.

• Extracts and synthesizes information from technical PDFs and academic research papers using Jina for deep architectural or theoretical investigation.

• Implements a tiered retrieval workflow, starting with narrow, reliable sources before expanding to broader web contexts to ensure high-precision results.

• Users should explicitly invoke mcp-research to trigger the research-heavy workflow when model uncertainty is detected.

• Always verify dependency versions against current registry data before suggesting upgrades or migration paths.

• When multiple sources conflict, the agent will report the discrepancy and prioritize the safest path, such as suggesting isolated testing or version pinning.

• Input requires specific search queries including library names, feature sets, and version numbers to reduce noise.

• Output provides synthesized findings with clear separation between sourced facts, verified documentation, and model-based inferences.

• Requires configured MCP server access to Context7, Exa, and Jina tools to function; if these services are unavailable, the skill performance will be constrained to standard model knowledge.