The standard that plugs AI into your tools was built for a human clicking 'allow'. Its new roadmap admits the caller is now a machine.

The Model Context Protocol — the open standard, started at Anthropic and now governed by the Linux Foundation's Agentic AI Foundation, that lets assistants call outside tools — published a new roadmap on 22 August. The bluntest section is about identity. MCP's authorisation model, the maintainers write, assumes a person with a browser at consent time, while the caller is increasingly a cloud workload with its own identity, acting for a user who is not present, or spawning sub-agents that should get narrower authority than their parent. Today's servers, the document says, lean on pasted API keys and long-lived refresh tokens. A new Agent Identity working group is forming to finalise DPoP proof-of-possession tokens and define delegation through workload identity federation and RFC 8693 token exchange. Four other priorities: one HTTP transport for both local and remote servers, spoken over stdin and stdout; a redesign of the tools/call return shape, which today allows two conflicting kinds of output at once; progressive discovery, so a client learns a server's tools as it needs them instead of swallowing the whole catalogue; and generating SDKs from the specification instead of maintaining them by hand.

✓ Verified · 2 sources

Related stories