Model Context Protocol 1.1 spec released with multi-server orchestration

Anthropic published the Model Context Protocol (MCP) 1.1 specification, adding multi-server orchestration, persistent session management, and a standardised resource and tool registration mechanism. The update moves MCP from a single-agent tool-calling interface toward a genuine inter-agent communication layer — allowing AI agents to discover, delegate to, and receive results from other specialised agents at runtime without hardcoded routing logic.

**Why this matters for enterprise AI architecture.** The dominant pattern for enterprise AI in production today is single-agent with tool calls: one LLM, a list of approved tools, and a human-in-the-loop checkpoint before any write action. MCP 1.1's multi-server orchestration enables a second pattern: agent networks where a planning agent decomposes a task, delegates subtasks to domain-specific agents (a document retrieval agent, a CRM write agent, a compliance checking agent), and aggregates results. This is the architecture required for complex workflows — contract review + CRM update + calendar booking from a single user request — that single-agent systems cannot reliably execute.

**Practical implications for APAC mid-market.** For enterprises currently deploying or evaluating AI, the 1.1 update signals that the agent-to-agent routing problem is being standardised at the protocol level, not solved ad hoc by each vendor. Teams building on top of Claude, or tools that have adopted MCP (including Cursor, Windsurf, and several enterprise workflow platforms), should expect the specification to stabilise rather than drift. That stability makes multi-agent architecture planning less risky than 12 months ago.

**Governance and control surface.** Multi-agent orchestration significantly increases the audit trail requirement. Each agent-to-agent delegation creates a new action scope. Enterprises implementing MCP-based architectures should build approval logging from the outset — not retrofit it after a runaway delegation chain causes a data access incident.

**AIMenta's editorial read.** MCP 1.1 is a credible foundation for production multi-agent systems. The protocol is maturing faster than most enterprise adoption curves, which creates a planning opportunity: organisations that build MCP-compatible agent architecture now will have significantly less integration work as the ecosystem consolidates.