Are MCPs Overhyped? A Rant about MCPs — Henry Mao, Smithery

Introduction to MCPs 00:03

• Henry Mao introduces himself as the founder of Smithery and a member of the MCP committee, aiming to provide an overview and critique of the MCP ecosystem.
• He shares his background, including his experience co-founding Jenny AI and his involvement with the ARC AGI challenge.

Challenges in the MCP Ecosystem 01:39

• Despite advancements in AI, autonomous agents are not yet solving everyday tasks, leading to a recognition of the limitations in the current AI landscape.
• The release of the model context protocol (MCP) by Anthropic aimed to standardize how models communicate with services but has introduced new challenges.

User Problems with MCPs 02:49

• Fragmentation: An increasing number of MCP servers makes it difficult for users to identify high-quality options.
• High friction installation: Complex installation processes can deter users, and security concerns arise from potentially insecure MCPs.
• Economic uncertainty: There are no clear plans for agent payments or subscription management for AI services.

Developer Problems with MCPs 03:50

• Hosting challenges persist, even with improved transport solutions; developers face issues like session management and resumability.
• Lack of tooling: Developers struggle with basic resources for testing and designing effective MCPs.
• Distribution and observability issues hinder the discovery and improvement of MCPs, alongside concerns about monetization.

Smithery's Role and Solutions 05:01

• Henry discusses the founding of Smithery to address these challenges and facilitate the growth of AI-native services.
• A demo of the Smithery playground showcases the potential of AI agents using curated MCPs to complete tasks, like creating tickets on linear from GitHub issues.

Future Perspectives 07:01

• The talk concludes with optimism for the MCP ecosystem, emphasizing the shift towards tool calls over clicks and the importance of enhancing the agent experience.
• Henry encourages collective efforts from the community to build the future of AI services.