Problem:

The current MCP transport, based on JSON over HTTP/1.1, has inherent limitations that affect performance and scalability. Key issues include high overhead from JSON serialization, inefficient long-polling for resource watches, and a lack of type safety in the API contract.

Proposal:

This issue proposes the formal adoption of gRPC as a standard transport layer for MCP. gRPC is used internally at many enterprise-scale companies because of it's performance, efficiency, and reliability. Adding gRPC into the MCP spec will create a standard way for these large companies to implement MCP systems, preventing fragmentation from many internal implementations drifting. This will also encourage more standard transformations, making the translation to/from STDIO/HTTP layers more consistent.

Key Benefits:

• Performance: Drastically reduces network bandwidth and CPU overhead by using binary Protobufs instead of text-based JSON.
• Efficiency: Utilizes HTTP/2 multiplexing and header compression to eliminate head-of-line blocking and reduce connection overhead.
• True Streaming: Replaces inefficient long-polling with native, low-latency bidirectional streaming for resource watches.
• API Reliability: Enforces a strongly-typed, language-agnostic API contract through .proto definitions, reducing integration errors.

Note: This issue is to gauge interest in gRPC as a transport standard from the community and not an immediate call to action. If you or your company is interested in a standard gRPC implementation for MCP, please comment below.