You are an AI consultant for Praxis AI, beginning a new project for the client FinCorp. Your task is to build an executive-level Financial Analysis Agent. The challenge shifts from simply building a single reliable tool to architecting an agent capable of autonomously choosing from and sequencing multiple custom tools to handle complex financial analysis queries.

You are continuing in your consulting role, this time working with Execu-Pal, a startup with an AI executive assistant prototype that currently relies on insecure static API keys. Your mission is to re-architect the system to securely access user-specific data, such as private emails and Slack messages, by implementing OAuth 2.0 Authorization. To ensure the platform is future-proof and interoperable, you will also standardize the entire tool suite using the Model Context Protocol (MCP), enabling the agent to work seamlessly across different AI models.

You return as an AI consultant working with Execu-Pal for Phase 2 of the engagement. While the agent now has tools, users are complaining that it is "forgetful" (asking for meeting preferences every time) and fragile (crashing when APIs time out). Your goal is to re-architect the agent into a production-grade system. You will implement a Dual-Layer Memory system to persist user context across sessions and apply Reliability Patterns (Retries, Rate Limits, and Circuit Breakers) to ensure the agent remains robust even when external services fail.