Martina Zrnec Launches Stacky: Bridging Content Hubs and AI Assistants via MCP

The Kid in the Candy Store Problem

Martina Zrnec has launched Stacky, a Model Context Protocol (MCP) server designed for the content curation platform Stacklist. The system enables AI assistants to autonomously manage stacks and cards for a user base of 14,000 people.

Why This Matters

The technical reality of building AI-integrated tools like Stacky shifts the engineering focus from pure logic to semantic communication. Developers must prioritize tool descriptions as onboarding documents for non-human readers, ensuring that input schemas and pagination are specifically optimized for a model’s context budget rather than human consumption.

Key Insights

• The Model Context Protocol (MCP) acts as a standardized interface for AI assistants to interact with external tools like Google Drive or Stacklist (2026).
• Writing precise tool descriptions for AI models is a major engineering bottleneck, functioning more as onboarding documentation than traditional API specs.
• Stacky enables AI hands to perform complex tasks such as auto-tagging content and generating summaries across entire curated collections.
• Robust MCP implementation requires auth flows that handle token rotation and error messages that specifically inform the model how to correct its state.
• Pagination strategies in AI-driven tools must be designed to respect the context budget of the model to maintain operational reliability.

Practical Applications

• Use Case: Automating content hub growth by allowing AI to add cards from URLs and generate knowledge capsules from markdown. Pitfall: Using vague tool descriptions that lead to the model calling functions with incorrect or malformed parameters.
• Use Case: Implementing cross-collection search and summary generation within Stacklist hubs. Pitfall: Implementing lazy rate limiting that degrades user experience during intensive AI-driven curation tasks.

References:

• https://dev.to/mz06/we-talked-about-this-for-two-years-now-you-can-talk-to-it-2di0