MCP servers for mobile developers: the good, the bad, and the “sorry, you ran out of context”

Model Context Protocol (MCP) is maturing fast. What started as a promising way to connect models to tools and resources is now evolving into a broader ecosystem with real applications, sharper security considerations, and increasing overlap with agent-based systems. For mobile developers, this raises an important question: how does MCP actually fit into mobile apps and mobile-centric architectures?

This talk focuses on MCP through the lens of a mobile developer, assuming basic familiarity with the protocol. Instead of re-explaining what MCP is, we dive into what changes once MCP is used in real systems that have limited context budgets, strict security boundaries, and long-lived user sessions.

Using concrete examples, we explore:
• How tools and resources can silently consume context and why this matters more on mobile
• Common failure modes when MCP is introduced into existing app and backend flows
• Where prompt injection and unintended tool behavior show up in practice
• What recent work around safe code execution, sandboxing, and MCP-based apps means for mobile teams

We also look at recent developments in the MCP ecosystem:
• The shift from experimental servers to app-like MCP deployments
• Growing alignment with agentic AI standards and foundations
• How this affects long-term maintainability, ownership, and trust boundaries

Finally, we zoom out. There is a real possibility that part of the future app landscape will consist of MCP-enabled apps, or apps that expose carefully scoped capabilities via MCP rather than traditional APIs alone. If that future materializes, mobile developers are in a strong position to take advantage of it. We already think in terms of capabilities, permissions, user context, lifecycle management, and secure client-server boundaries, all concepts that map surprisingly well onto MCP’s model.

Rather than presenting MCP as inevitable, this talk treats it as a developing shift worth understanding early. It helps mobile developers reason about when MCP might make sense, when it probably does not, and how to prepare for a future where apps, tools, and models increasingly meet in the middle.

Attendees will leave with practical mental models, concrete pitfalls to avoid, and a clearer sense of how MCP could fit into modern mobile development workflows today and tomorrow.