Descriptions:
This Latent Space podcast episode marks the one-year anniversary of Anthropic’s Model Context Protocol (MCP), featuring an in-studio conversation with David Soria Parria, the protocol’s creator. Soria Parria traces MCP’s evolution from a local-only tool for Claude Desktop into a remote-capable, enterprise-grade standard with OAuth-based authentication—work developed with the help of OAuth experts brought in between March and June 2025. He identifies the April 2025 moment when Sam Altman, Satya Nadella, and Sundar Pichai publicly committed Microsoft, Google, and OpenAI to MCP adoption as the protocol’s defining inflection point.
The episode also serves as the announcement context for the Agent AI Foundation (AAIF), a new open-source governance body jointly stewarded by Anthropic, OpenAI, and Block. Soria Parria explains why the foundation deliberately launched with a narrow scope—starting with MCP rather than attempting to standardize payment protocols or other adjacent specs—and why getting adoption right before expanding governance was a priority. Leaders from OpenAI, Goose, and the Linux Foundation join the second half to discuss the foundation’s structure.
On the technical side, the conversation covers the latest protocol iteration adding long-running task support (enabling deep research workflows and agent-to-agent communication), the persistent underuse of MCP’s sampling feature due to sparse client support, and the practical differences between local and remote MCP server deployment. A community hackathon in New York is announced for early 2026. For developers building on MCP or tracking AI interoperability standards, this episode offers unmatched first-person depth from the protocol’s original author.
📺 Source: Latent Space · Published December 28, 2025
🏷️ Format: Podcast
