> Markdown version of [/videos/100305-api-mcp-or-mcp-app-choosing-the-right-surface-for-ai-agents?t=1186](https://www.wearedevelopers.com/videos/100305-api-mcp-or-mcp-app-choosing-the-right-surface-for-ai-agents?t=1186). Every page supports `.md` or `Accept: text/markdown`. Links point to the HTML versions so they work for humans too. Agent guide: [/agents.md](https://www.wearedevelopers.com/agents.md).

---

# API, MCP or MCP App? Choosing the right surface for AI agents

Are standard APIs enough when your end user is an AI? Learn how the Model Context Protocol shifts agency directly to the model through dynamic apps.

- **Speakers:** [Rishabh Budhiraja](https://www.wearedevelopers.com/@rishabh-budhiraja), [Rishi Deorukhkar](https://www.wearedevelopers.com/@rishi-deorukhkar)
- **Event:** World Congress 2026 Europe
- **Published:** July 10, 2026
- **Duration:** 29:59
- **URL:** https://www.wearedevelopers.com/videos/100305-api-mcp-or-mcp-app-choosing-the-right-surface-for-ai-agents

## Summary

As customer journeys increasingly consolidate into AI agent interfaces rather than traditional web sessions, simply building an API is no longer enough when the end user is not a human developer. Engineering for this new paradigm requires understanding the vital distinction between standard APIs, the Model Context Protocol (MCP), and MCP apps. While APIs dictate fixed developer workflows, MCP shifts agency directly to the model through rich, semantic descriptions. Taking this further, MCP apps blend backend functionality with dynamic, interactive UI widgets sandboxed right inside host agents like ChatGPT and Claude. This evolution fundamentally changes how user interactivity bridges the gap between complex actions—such as discovering tailored product offerings or evaluating rich price histories—and live LLM conversational flows.

Building and evaluating MCP apps introduces distinct engineering disciplines often absent from their original backend counterparts. Most notably, tool descriptions function as mission-critical prompts rather than basic documentation; a single rephrased line can dictate whether an agent successfully leverages a capability or wildly hallucinates arguments. For this reason, conventional unit tests constantly fall short. Teams must adopt robust evaluation protocols using frameworks like Promptfoo to benchmark diverse models against varied, real-world user queries and validate exactly which tools are invoked. Further minimizing friction is the realization that dynamic UI widget states push functional context directly back to the model. An interface action—such as selecting a specific product in a carousel—updates the AI's understanding seamlessly, setting up highly personalized subsequent dialogue without redundant prompts.

To handle fragmentation across rapidly emerging agent platforms, leveraging cross-platform tools like the mcp-use framework allows developers to write business logic once and deploy unified interactive connectors to multiple frontier models. This strategy seamlessly aligns with a necessary mindset shift from merely servicing rote searches to designing full, contextual agent journeys. By shaping exactly how these advanced tools are exposed and carefully evaluating the end-to-end interface flow, engineering teams can build robust AI storefronts that systematically reduce friction, smoothly reject prompt injections with strong validation, and naturally adapt to shifting AI search optimization paradigms.

**Keywords:** model context protocol, MCP app development, LLM tool invocation, semantic tool descriptions, promptfoo evaluation, agent journey design, cross-platform MCP frameworks, dynamic client registration, interactive UI connectors, AI search optimization, widget state context, API replacement strategy, prompt injection guardrails, e-commerce AI integration

## Chapters

1. **Core concepts and advantages of the Model Context Protocol** (00:34) — How MCP shifts agency from developers to models by enabling dynamic API discovery and structured tool calls.
1. **Architectural overview of the trial Edialo MCP implementation** (03:11) — Offering rich e-commerce search capabilities through trial MCP tools like product history mapping and custom integrations.
1. **Securing connections using dynamic client registration and metadata** (05:35) — Managing secure AI agent access through client metadata documentation rather than decentralized per-server registration.
1. **Evolving AI responses from text walls to interactive interfaces** (07:47) — The transition from text-only output to embedded interactive UI widgets for richer product exploration.
1. **Managing context and interaction state in sandboxed widget frames** (10:55) — How sandboxed MCP user interface elements synchronize client interaction state back to the model context.
1. **Managing cross-platform deployment complexities for AI agent applications** (14:36) — Using unified frameworks to abstract API differences between separate provider capabilities and simplify cross-platform deployment.
1. **Optimizing semantic tool descriptions for reliable model invocation** (15:58) — Crafting precise documentation prompts to control how and when large language models select available functional tools.
1. **Integrating user interfaces and holistic journeys into agent context** (18:04) — Shifting application design to treat frontend interactions as continuous contextual inputs for agent-driven task completion.
1. **Running automated evaluation tests for conversational tool selection** (19:46) — Validating AI component reliability and prompt effectiveness against explicit test conditions using open-source evaluation runners.
1. **Visualizing the future of proactive multi-tool agent orchestration** (23:53) — Exploring capabilities where language models seamlessly string together multiple personalized applications to solve complex user tasks.
1. **Q&A on platform discovery, security, and prompt injection** (25:02) — Resolving challenges related to AI engine optimization, protecting endpoints from malicious payloads, and tracking framework updates.

## Related Moments

- [Introduction to MCP apps and dynamic UI elements](https://www.wearedevelopers.com/videos/1981-from-zero-to-mcp-ui-hero-building-context-aware-ai-apps-that-actually-work) (from "From Zero to MCP UI Hero: Building Context-Aware AI Apps That Actually Work")
- [Integrating Model Context Protocol for non-technical users](https://www.wearedevelopers.com/videos/100091-3-ways-to-rebuild-the-data-stack-for-agents) (from "3 Ways to Rebuild the Data Stack for Agents")
- [Standardizing agent interactions with the Web MCP proposal](https://www.wearedevelopers.com/videos/1805-wearedevelopers-live-modern-devops-for-iot-devices-and-more) (from "WeAreDevelopers LIVE - Modern DevOps for IoT Devices and More")
- [Treating AI tools as standard governed enterprise capabilities](https://www.wearedevelopers.com/videos/100336-from-apis-to-mcp-enterprise-governance-registry-and-controls) (from "From APIs to MCP: Enterprise Governance, Registry, and Controls")
- [Summarizing developer experience and artificial intelligence companions](https://www.wearedevelopers.com/videos/884-forget-developer-platforms-think-developer-productivity) (from "Forget Developer Platforms, Think Developer Productivity!")
- [Applying context engineering across the full software lifecycle](https://www.wearedevelopers.com/videos/100256-can-this-elephant-dance-ibm-bob-and-the-future-of-ai-first-software-development) (from "Can This Elephant Dance? IBM Bob and the Future of AI-First Software Development")

## Related Articles

- [Everything a Developer Needs to Know About MCP with Neo4j](https://www.wearedevelopers.com/magazine/604-everything-a-developer-needs-to-know-about-mcp-with-neo4j)
- [WebMCP: Empowering Agents as First-Class Citizens of the Web](https://www.wearedevelopers.com/magazine/696-webmcp-empowering-agents-as-first-class-citizens-of-the-web)
- [WWC24 Talk - Scott Hanselman - AI: Superhero or Supervillain?](https://www.wearedevelopers.com/magazine/469-wwc24-talk-scott-hanselman-ai-superhero-or-supervillain)
- [What is Agentic Programming and Why Should Developers Care?](https://www.wearedevelopers.com/magazine/625-what-is-agentic-programming-and-why-should-developers-care)

## Related Jobs

- [Senior Backend Developer — AI: MCP & Agent Engine](https://www.wearedevelopers.com/jobs/48297-senior-backend-developer-ai-mcp-agent-engine) at **basebox GmbH**
- [Principal Product Manager, Agent Platform](https://www.wearedevelopers.com/jobs/ext/277541-principal-product-manager-agent-platform) at **GitHub**
- [Principal Field Architect - AI Agents](https://www.wearedevelopers.com/jobs/ext/1442858-principal-field-architect-ai-agents) at **Twilio**
- [AI Full Stack Engineer](https://www.wearedevelopers.com/jobs/ext/1354435-ai-full-stack-engineer) at **Almedia**
- [Staff Software Engineer, Copilot Experiences](https://www.wearedevelopers.com/jobs/ext/164361-staff-software-engineer-copilot-experiences) at **GitHub**
- [AI Software Engineer (Germany)](https://www.wearedevelopers.com/jobs/48317-ai-software-engineer-germany) at **Sunhat**