Ralph-Loop-Ultra

The mcp-tool-manager is a sophisticated utility designed for autonomous agents to manage service dependencies in complex development workflows. It serves as an abstraction layer that enables developers and AI agents to switch between production-ready live APIs and sandboxed mock environments without modifying the core application logic. This skill is essential for maintaining high velocity during CI/CD processes, local debugging, and integration testing, especially when external services like Stripe, AWS, or Twilio are rate-limited, prohibitively expensive, or unavailable.

The skill standardizes how an agent interacts with external systems by utilizing the Model Context Protocol (MCP). It provides a structured mechanism to define configurations for both 'live' and 'mock' modes, allowing the agent to dynamically load the appropriate JSON-based tool definitions based on the current environment variable. Furthermore, it supports an intelligent auto-generation workflow, where the agent can consume OpenAPI specifications to synthesize mock tools on-the-fly if a predefined mock is missing, ensuring minimal friction when onboarding new external dependencies.

• Enables seamless hot-swapping between real-world API endpoints and simulated mock toolsets.

• Integrates directly with the Claude Code ecosystem through MCP to extend agentic capabilities.

• Provides automated mock generation workflows using OpenAPI/Swagger specifications to reduce manual boilerplate.

• Supports environment-aware configuration via environment variables like RALPH_MCP_MODE.

• Facilitates robust testing patterns by allowing agents to execute isolated unit and integration tests.

• Use this skill when working on projects that heavily rely on third-party SaaS platforms to prevent accidental production impact.

• Ensure that all mock tools are defined within the .ralph/ directory using standard JSON structures compatible with MCP specs.

• Maintain parity between live and mock schemas to ensure that self-correction loops remain effective during the transition from testing to deployment.

• Configure the CLAUDE.md file to map service aliases to their respective live or mock endpoints.

• The agent expects environment variables to be set correctly to toggle between production and development modes; verify these prior to running long-cycle autonomous tasks.