Common themes emerged among the 114 speeches at the conference, including:
“MCP Is Not Dead”: Late last year, social media debated whether MCP was dead because applications could use the command line interface (CLI) instead of MCP to communicate with agents. Every speaker addressed this meme directly, and the Maintainers Roundtable noted that both MCP and CLI are an alternative way to work with agents.
MCP reached 110M+ SDK downloads/month and its biggest usage week during the peak “MCP is dead” conversation. Amazon, Uber, Duolingo, Nordstrom, Bloomberg have all shared information about effective manufacturing deployments.
- Gateway + Registry = Business Pattern: Amazon, Uber, Docker, Nordstrom, and Obot are all defined using the same agent architecture: a designated MCP server catalog, a central registry for discovery/compliance, and an MCP gateway for authentication, policy enforcement, and audit logging.
- Context Bloat Resolvable: With continuous tool discovery, tool searching, deferred loading, and capabilities such as “meta-skills” that dynamically load the context, it is possible to reduce the LLM context window usage from 22% to almost zero percent.
- Skills associated with MCP: Skills, MCP, and context files are complementary elements in an agent configuration rather than competing elements. Capabilities manage context and include domain information while MCP provides access to tools and transport.
- Security is a Top Business Concern: Including Identity management, data exfiltration, supply chain attacks, escalation of permissions, and the “lethal trifecta” (private data + untrusted content + external comms), businesses worry about deploying MCP without adequate security controls.
- Background/Future Strong Agents: Presenters predict that in the future agents will work longer hours, wake up on schedules, and work independently. MCP work primitives enable async/long-running operations with state machines and notifications.
- First agent design: The presenters recommend that you design your systems to be agent-friendly, not just human-friendly. Don’t just wrap APIs – design intent-based tools that target agent buyers.
- MCP Apps = The New Web? MCP applications are interactive GUIs that are rendered inline as iFrames within agent dialogs. VS Code, Cursor, Copilot, ChatGPT, and Claude support them as a way to break AI applications into composable UI panels for users.
- Testing is a Bottleneck: The guidance is to develop test programs before introducing production agents to ensure efficacy and reliability. If we make test tools available through MCP servers agents can improve themselves.
- Use Git as a proxy: Otherwise known as the “MintMCP pattern” – use Git repos to store agent status, progress tracking, and performance data. This leads to testability, reproducibility, and collaboration (branches/PRs) of self-learning agents that update their .md files).
