There is a contradiction here. Your governance context defines MCP as your proprietary product, Liminality MCP, a question-decomposition and routing server. The retrieved facts describe the Model Context Protocol, an open standard from Anthropic. You are asking for news on the MCP protocol while your internal context treats MCP as your tool. The facts do not contain any news or technical details about your proprietary Liminality MCP product. They only cover the open-source Model Context Protocol.
I must clarify: are you asking about the open-source Model Context Protocol or your proprietary Liminality MCP? Given the retrieved facts are exclusively about the open-source standard, I will answer based on those facts. If you intended to ask about your proprietary Liminality MCP, please provide specific details or context for that product. It is not present in the retrieved data.
### Answer Based on Retrieved Facts (Open-Source Model Context Protocol)
#### 1. Latest News and Updates
The Model Context Protocol is undergoing its largest update since launch. It is scheduled for release on July 28, 2026. Key changes include:
* **Stateless Architecture:** The update removes the initialize handshake and session ID requirements. This enables standard load balancing. * **Security & Extensions:** It introduces a formalized extensions framework, mandatory request headers, and hardened OAuth/OIDC authentication. * **Deprecations:** The features for Roots, Sampling, and Logging are being formally deprecated. These will remain functional until mid-2027 under a 12-month lifecycle policy. * **Compatibility:** Existing servers built on the 2025-11-25 specification will remain functional during the transition.
#### 2. How It Works for AI Agents
The MCP specification defines how clients, such as AI-powered IDEs or agents, and servers, data providers, communicate using JSON-RPC 2.0.
* **Integration Mechanism:** * **Tools Capability:** For tool-related operations, servers must support the tools capability. * **Endpoints:** * `tools/list`: Used by the client to discover available tools. * `tools/call`: Used by the client to execute a specific tool. * **Tool Definition:** Each tool requires a unique name and an inputSchema defined as JSON Schema.
* **Example Interaction:** An AI agent invokes a tool by sending a JSON-RPC request. For example, to get the weather in New York:
```json { "jsonrpc": "2.0", "id": 2, "method": "tools/call", "params": { "name": "get_weather", "arguments": { "location": "New York" } } } ```
* **Documentation:** The authoritative specification is available at modelcontextprotocol.io. You can also search the full spec via modelcontextprotocol.io/llms.txt or use an unofficial MCP documentation server.
---
### Pending Action / Clarification Request
The retrieved facts do not contain information about your proprietary "Liminality MCP".
* If you need news or technical details about Liminality MCP, your question-decomposition and routing server, please provide the relevant context or documents. This information is MISSING from the retrieved facts. * If you are satisfied with the information regarding the open-source Model Context Protocol, no further action is needed.
This answer was worked out by Liminality — Physea's engine that decomposes a request, grounds each part to a real tool, and returns a reusable, checkable route.
Connect it over MCP: https://mcp.physea.ai/mcp · physea.ai