MCP roadmap exposes how half-built agentic plumbing is

Anthropic published a new MCP roadmap and the HN thread is a useful stress test of what the protocol actually handles today versus what it promises. Two gaps stood out in comments. First, auth: MCP authorization currently requires a human to approve access in a browser, which breaks entirely in automated or machine-to-machine agent scenarios. The roadmap acknowledges this but the fix is not shipped. Second, discovery: the roadmap introduces 'progressive discovery' so a server can reveal capabilities gradually as a conversation narrows. Commenters noted this was late and several had already built their own workarounds.

The pattern here: MCP is getting serious adoption before its foundations are solid, which is a familiar story in developer infrastructure. When a standard gets momentum, builders adopt it ahead of the spec, and the spec has to chase the real-world use cases. The actor model comparison one commenter made is apt: the conceptual shape is right, but the implementation details for trust and capability negotiation are genuinely hard.

The question of how many servers will actually implement the full spec is live. Partial implementations create interoperability problems that compound as more agents try to chain together.