Summary of Key Changes
The latest version of the MCP (Model Context Protocol) (dated 2026-07-28) represents a significant shift from a stateful to a stateless design. This change, although seemingly minor on the surface, marks an important evolutionary step for Agents—smart intermediaries—from being simple tools for task execution to becoming entities capable of managing enterprise-level responsibilities. Previously, information such as status, permissions, and task progress was hidden behind persistent connections (sessions). Now, these details must be explicitly defined, passed between systems, and stored to address critical issues when Agents perform irreversible operations. By doing so, the protocol establishes a clear chain of accountability, ensuring that every action can be traced and someone is responsible for its outcome.
Detailed Explanation of the Changes
1. Statelessness Does Not Mean Forgetting
In the past, MCP operated like a continuous telephone conversation where the client and server maintained a connection, with the server remembering who the user was and where the previous discussion had stopped. However, this approach is fragile in enterprise-level applications: a server crash or sudden increase in traffic could cause ongoing tasks to fail. The new version of MCP requires each request to explicitly identify the user, the protocol version, and any previous context (e.g., the ID of a purchase task). It’s like going to the bank; you don’t need to occupy a window the whole time—you just bring the previous receipt the next time you visit.
Key Points:
- Status is no longer hidden within the connection but is represented as transferable references (e.g., contract draft IDs, task numbers) that must be provided with each new request.
- A multi-request tracking mechanism (MRTR) has been introduced. For example, if an Agent encounters an issue that exceeds the budget during a purchase, it will pause and ask for confirmation before proceeding.
- Repeated requests are prevented from causing duplication (e.g., the same order is not submitted multiple times, ensuring idempotence).
In other words, statelessness means turning vague, hidden information into clear, accountable processes.
2. MCP’s Simplification Leads to a More Dynamic Ecosystem
MCP was initially designed to include all Agent functions (such as task management and logging), but now it focuses only on the most basic tool interactions (like database queries and calendar creation). Other tasks and logs are handled by separate extensions. This separation is similar to how smartphones have dedicated apps for tasks like photography and payments.
Ecosystem Changes:
- A2A Protocols: Facilitate collaboration between Agents, such as a sales Agent handing over tax-related tasks to a specialized tax Agent.
- WebMCP: Allows web pages to directly interact with Agents, enabling them to perform actions without needing emulation (e.g., directly checking order details).
- Interface Protocols: Show users what the Agent is doing (e.g., displaying “checking inventory” on the user interface when an Agent uses a tool).
- Business Protocols: Convert natural language commands (e.g., “buy a monitor”) into structured orders with specific requirements (budget, brand, payment method).
In the future, the world of Agents will consist of a stack of protocols, each handling different aspects of business processes. No single protocol will cover everything, but they will work together seamlessly.
3. Making Agents Useful Requires Clarifying Responsibilities
It’s easy for Agents to perform simple tasks like checking the weather or reading documents, but more complex actions (deleting data, placing orders, signing contracts) pose challenges because they involve determining who the Agent represents and what permissions it has. Core issues include:
- Permission Security: Preventing malicious commands from causing data loss (e.g., by restricting Agents to only read data).
- Authorization Tracking: Every action must be recorded (who authorized it, when, and with what credentials).
- Business Requirements: Specific details like budgets and return policies must be defined in structured protocols.
In summary, for Agents to play a central role in business processes, it’s essential to clarify their authority: the system and users must understand what they can do and who is responsible for their actions.
4. The Registry Evolves from a Tool Directory to a Trust Gate
The Registry currently acts like a directory of available MCP tools. In the future, it will become a platform with strict validation processes:
- Trust Management: Verifies the identity of tool developers and checks for malicious code.
- Enterprise Filtering: Enterprises can set rules (e.g., only allowing access to approved financial tools).
- Value Creation: Tool developers need to offer unique business value (e.g., by integrating company-specific policies and error correction data).
For enterprises, the Registry acts as a gatekeeper, ensuring they use secure and reliable tools. For developers, it’s a competitive landscape where the focus is on providing deep business functionality rather than just having compatible interfaces.
5. Challenges for Implementing Agents in China
Chinese companies face significant system fragmentation (various identity systems and approval processes). While MCP solves the connection issues, additional challenges need to be addressed:
- Cross-system Identity Management: How to automatically revoke Agent permissions when employees leave a company?
- Permission Assignment Based on Actions: How to adjust permissions as tasks progress through different departments (sales, legal, finance)?
- Resuming Long-running Tasks: Can interrupted tasks (e.g., purchases) be resumed after a system crash?
- Emergency Handover: How to efficiently transfer tasks to human operators when Agents encounter issues?
These challenges vary by industry and require tailored solutions. Only by overcoming these hurdles can Agents move from being demonstration tools to valuable, practical components of business operations.
Conclusion
The stateless update in MCP is not just a technical advancement; it’s a necessary step for Agents to become truly useful in real-world applications. It drives the entire ecosystem to shift from focusing on tool connectivity to emphasizing responsibility management. In the future world of Agents, the key will not be which protocol is the most powerful but who can establish a clear chain of accountability. This will enable enterprises to confidently delegate tasks to Agents. For developers and entrepreneurs, the opportunity lies in solving practical business problems, not just integrating protocols. For users, the value of Agents lies in their reliability—ensuring that they perform their tasks correctly and that someone is accountable for any issues that arise.