Model Context Protocol

On ToTheMoon, Model Context Protocol is the answer to how a chat turns into an agent. The concept arrived in June 2025 in two special episodes on agents without code: a plain chat replies in text, while an agent can open a folder, sort files and assemble a presentation, and the new level begins not with a prettier answer but with the model's access to actions through tools and the MCP protocol (episode extra06). The hosts read MCP as a connecting layer that makes tools reusable and loosens the grip of any single model vendor.

In episode extra07 the hosts showed how to connect MCP servers and drew the boundary at once: give the agent only the access the scenario needs, and keep critical actions behind human confirmation. When OpenAI allowed third-party MCP servers, the channel showed how Deep Research becomes a construction kit, but noted this is the iPhone-versus-Android trade-off where everything is configured by hand (episode 061). In December the hosts observed that the protocol had been handed to the Agentic AI Foundation as a public good (episode 088).

By 2026 MCP had become a marker of maturity for the channel. Episode 131 describes MCP servers as connectors to internal resources such as Confluence, Jira and the codebase, giving ChatGPT context without crossing access boundaries. The hosts call MCP connectors a transitional mechanism (episode 146). The summary view comes in episode 154: over the podcast's lifetime the base model changed less than the wrapping around it, from prompts to APIs to MCP to sub-agents, and that wrapping changed the situation radically.