What problem does it solve? Connecting a new external tool to an AI coding agent via MCP often involves guessing at configuration JSON, trusting unverified community servers, or silently overwriting existing settings. This Skill enforces a research-first, confirmation-gated process so integrations are accurate, disclosed, and reversible. ## Core Features & Use Cases - Verified research: Searches official vendor documentation and repositories for the canonical MCP server, labeling community servers and flagging security implications before any change. - Plan-and-confirm gate: Shows the exact mcpServers JSON entry, install command, environment variables, and affected files, then stops until the user explicitly approves. - Surgical application: Merges only the named server entry into .claude/mcp.json, preserves unrelated keys, adds one documentation bullet under ### MCP Servers in CLAUDE.md, and verifies no literal secrets were written. - Use Case: You want to connect a ticket tracker to your agent. The Skill researches the official MCP server, shows you the exact config diff, and only after your approval writes the entry and optionally sets ticket_tracker in answers.env. ## Quick Start Ask the agent to integrate a specific external tool through MCP, for example by naming the ticket tracker or service you want connected.
