AI Investing, Engineered: Strands Agents with MCP & ACP for Personalised Insights

When we began developing Advisory [link], our vision for a Financial and Investment Agent went far beyond dashboards and automated alerts. From the start, we focused on modularity, scalability, autonomous intelligence and rich user interactions - allowing the platform to adapt to different market segments through a cost-efficient approach.

The Advisory stack consists of a microservice architecture that integrates Strands AI Agents, a Model Context Protocol (MCP), an Agent Communication Protocol (ACP), a financial API server, and a Firebase-based personalised push notification system.

Strands Agents: Intelligent Conversational Advisors

Built by using the Strands Agents SDK, Advisory offers autonomous reasoning, tool invocation, and natural-language interactions. This foundation enables personalised, real-time trading insights with minimal engineering overhead.

The system leverages Python-based declarative tooling to seamlessly integrate multiple AI models and APIs. We dynamically route across LLMs we’ve rigorously tested to ensure optimal quality and cost efficiency within each user plan.

Currently, Advisory uses gpt-5-mini for user conversations with an estimated processing cost of $0.024 per user  daily, or approximately $0.8 monthly. The agent analysis and data preprocessing cost is estimated to be at around $0.05 per day for each stock.

MCP Servers for Real-Time Agent Decisions

The communication relies on the following API by using MCP (Model Context Protocol):

• Financial Data API: Implements endpoints for live exchange rates, technical indicators, price predictions, and AI insights with input validation and rate limiting for robust performance.

• User Profile: Integrates with the database to fetch user-specific portfolios, risk preferences, and asset favourites and previous chat conversation enabling high personalization.

• Notification Monitoring: Continuously monitors asset rate changes for users and triggers push notifications when there is actionable Investment data. We also monitor users' engagement on each notification

• Push Notification: During the analysis, the agent reviews the users' interests, determines which users should receive the relevant updates and triggers sending a notification for the appropriate users. This reduces bombarding users with unnecessary data. We chose Firebase for insights delivery and real-time user analytics that is looped back into the user behaviour profile update.

Platform Highlights: Microservices & Docker

Advisory’s stack is microservices all the way: every component (agents, MCP service, notification handler) is containerized with Docker and communicates via secure APIs. This ensures rapid iteration, elastic scaling under load, and consistently low latency on both Android and iOS. Key advantages include:

• Independent Service Scaling and Updates: Each microservice can be upgraded or scaled without affecting others.

• Robust Development and Deployment Pipelines: Local and cloud environments mirror each other, facilitating continuous delivery.

• High Availability and Observability: Services are instrumented with logging and monitoring, critical for financial app reliability.

Conclusion

By orchestrating Strands agentic interface, a comprehensive MCP framework, real-time financial and market insights, personalised user behavior metadata in the agent loop, we established a scalable, intelligent, and user-focused cost efficient AI stack.

For the success of any AI stack, resource and per-user cost optimization are crucial from the earliest stages of SW design. This modular, Dockerized architecture not only accelerates development but also enhances user engagement through proactive AI-driven interactions - all while maintaining predictable and sustainable operational costs.

Link to the Advisory on Play Store