Implementing MCP in Sarion

Introduction

In 2024, a fundamental concept began to emerge for the future of artificial intelligence: the ability for models to move beyond their pre-trained limits and interact with real systems, data, and applications.

The Model Context Protocol (MCP), initially introduced by Anthropic and then rapidly adopted by OpenAI and other industry leaders, represents this exact shift: a common language that enables AI assistants to use tools, access information, and perform actions in the real world.

Within Sarion — the productivity application that integrates task management, notes, journaling, and an AI assistant — adopting MCP opens the path to seamless collaboration between users and intelligent models, without locking into a single AI vendor or interface.

What is the Model Context Protocol?

MCP is an open standard designed to simplify how AI systems integrate with external services.

Before MCP, every tool required custom connectors for each AI model — a costly and fragmented approach. MCP eliminates this by defining a shared protocol.

AI no longer works in isolation: it can read up-to-date documents, query databases, create notes, manage tasks, and automate workflows.

Thanks to this approach, any application that exposes tools through an MCP server can be used immediately by compatible AI models: Claude, ChatGPT, Gemini, Cursor, AI-first IDEs, productivity environments, and more.

Why AI Needs MCP

Large language models, no matter how advanced, have a clear limitation:

They do not know what happened after their training, and they cannot act directly inside a user’s environment.

MCP solves this in three ways:

1. Access to Real, Updated Data
2. The AI can query calendars, projects, documents, databases, and personal tools.
3. Standardized Integrations
4. If a service exposes tools via MCP, any compatible AI can use them.
5. Enabling Autonomous AI Agents
6. The assistant can reason, make decisions, and take action — not just respond.

This means a request like:

“Organize my marketing project schedule and create tasks ordered by priority.”

doesn’t just result in a text suggestion — it becomes real tasks created in the user’s application.

How MCP is Already Being Used Today

Adoption has been fast and significant:

• Notion allows AI models to read and update documents.
• GitHub exposes issues, PRs, and repositories for chat-driven development workflows.
• Zapier opens access to 30,000+ applications through AI-driven automations.
• Google ADK enables real-time web data scraping and analysis.
• n8n integrates MCP for complex automation workflows orchestrated by AI.

This is no longer science fiction — it’s the beginning of AI-first applications where chat becomes the primary interface.

Sarion Today: The Application

Sarion combines a clean, organized interface with an integrated AI assistant. The application includes:

• Task and list management
• Notes and ideas
• Personal journaling and mood tracking
• Folder-based organization
• Daily habits
• Conversational AI assistant (powered through Mastra AI and Gemini models)

The UI is divided into two areas:

• Left → Chat interface with the AI assistant
• Right → Traditional point-and-click productivity tools

MCP Integration in Sarion

The goal is to expose Sarion’s functions as AI-invokable tools, allowing users to interact with Sarion from any AI assistant — not just the web app.

Example:

“Claude, add a task ‘Send client proposal’ in the Work folder, due tomorrow.”

Claude → MCP → Sarion backend → Task is created.

No app opened.

No clicks.

No copy-paste.

Features that will be exposed through MCP:

• Create, update, and complete tasks
• Manage notes
• Create and query journal entries
• Track daily habits
• Organize folders and workspace

Security and Control

To ensure reliability and safety:

• OAuth or secure token authentication
• Granular scope permissions (read/write/selective)
• Confirmation for sensitive operations
• Rate limiting and misuse protection

The goal is to give the AI the ability to act — but always in a controlled and secure way.

Roadmap

• Define complete MCP tool schema
• Implement MCP server in Node.js
• Deploy as a serverless function on Vercel
• Test integration with desktop AI assistants (Claude, Cursor, ChatGPT)

Conclusion

MCP represents a shift in perspective: applications are no longer isolated systems, but components of a collaborative AI ecosystem.

Sarion can become not just an app, but a tool accessible to any AI — empowering users to always get the most effective and intelligent support possible.