Beyond JSON-RPC: Scaling Model Context Protocols With gRPC in the Py...
summary
Proposal adds SEP-1352 to make MCP use Protobuf over gRPC as a pluggable transport, claiming binary encoding cuts an 8 MB context from ~15–20 MB in JSON down to ~8 MB (numeric payloads to ~4 MB) and removes JSON parse CPU overhead. Migration is a simple swap of Uvicorn for a serve_grpc entry with identical tool logic, supports mTLS/service meshes and polyglot SDKs, and a demo shows 0.7s vs 2s for 1,000 call_tool roundtrips.
AI-written summary · about
chapters (14)
- 0:00gRPC and MCP scaling
- 1:03AI tool integration problems
- 2:06Model Context Protocol solution
- 3:28MCP with JSON-RPC state
- 4:51JSON-RPC HTTP limitations
- 6:53gRPC integration proposal
- 7:35MCP over gRPC architecture
- 9:02gRPC security and auth
- 10:25MCP server migration
- 12:29Client migration to gRPC
- 13:55gRPC advantages for MCP
- 14:38Roadmap and ecosystem
- 16:26Live demo gRPC vs HTTP
- 18:41Q&A
related talks
28mFrom Backend Powerhouses to Web-Friendly APIs: Unlocking
19mgRPC as the Native Transport for Agent Protocols (A2A and MCP)
18mgRPC and AI
14mOverview of gRPC
42mMCP in Practice: One App to Understand It All
26mDemystifying Protocol Buffers and gRPC
12mModel Context Protocol (MCP) and agentic skills
32mEnhancing the Performance Testing Process for gRPC Model Inferencing at Scale