FDE Times: DeepLearning.AI + Anthropic MCP course — learn it fast, then rebuild for the 2026-07-28 spec
A 5 November FDE Times review of Elie Schoppik’s ~2-hour Intermediate MCP course says the FDE value is the architecture (one connection per client; tools vs resources vs prompts; local then remote), then rewriting exercises against the July 2026 stateless MCP spec that deprecates sampling and protocol-level logging.
A 5 November FDE Times books-and-courses review covers DeepLearning.AI and Anthropic’s short course “MCP: Build Rich-Context AI Apps with Anthropic” (11 lessons, about two hours, Intermediate). The pitch for FDEs is not finishing the videos — it is connecting models to the customer systems they already run (tickets, documents, databases).
Keep: one MCP client connection per server (so a timed-out document library does not take ticket search with it); classifying discovery items as tools (actions), resources (read-only data) or prompts (reusable templates); and moving from a laptop stdio prototype to a remote Streamable HTTP server where auth and ops ownership show up.
Then rebuild: the 2026-07-28 MCP spec makes the protocol stateless, removes protocol-level sessions / the initialize handshake as described in the guide, and deprecates sampling and protocol-level logging. Interview evidence the review recommends is a GitHub remote server with a README that states the spec version and why each capability is a tool, resource or prompt.
Source: fdetimes.net