github.com/JeffreyRichter/MCP v0.0.0-20260412154919-fe1532555d96
LLM or AI-agent context phrasechmod +x (make executable)
SHA-2561b0ac8857d60d0e5c223e04aadfd4f746970f69ac7b7a073692a82a06d290e9f
MaleculeO(Er)H(F)
Evidence
94 - Existing MCP authorization solutions continue to work as-is since they already follow HTTP industry standards. And, it is easier to debug authorization issues with HTTP REST.
95 - Existing HTTP middleware services such as API Gateways, CDNs, SSL/TLS termination, front doors, metrics/logging collection, and distributed tracing can be used.
96 - The Streamable HTTP transport doesn’t allow this because the middleware service must crack open and parse the JSON-RPC in the HTTP request/response bodies to log methods like number of times each tool/prompt/resource got invoked, request failures, etc.
97- Improved performance by reducing client/server messages (such as list changed notifications) using HTTP's standard caching & optimistic concurrency patterns (etags).
98- Enables long-running and streaming-response tool calls allowing a tool call to act as an AI agent. Neither feature is supported by MCP today.
99- This SEP enables one device (say a PC) to initiate a tool call and another device (say a mobile phone) to track the state of the tool call and continue a conversation with an LLM once the tool call completes on any device.
100- The HTTP REST transport can be used securely for both local and remote MCP Servers (I have this working too). The need to use a stdio transport for local and Streamable HTTP for remote becomes unnecessary. Migrating code from a local MCP Server to a remote MCP Server (or vice versa) is trivial now. I also believe that the standard HTTP authentication can be made to work in this local mode which the current stdio transport doesn’t support at all.
7FROM mcr.microsoft.com/azurelinux/base/core:3.0
8COPY --from=builder /build/mcpsvr .
9RUN chmod +x mcpsvr
10EXPOSE 8080
11CMD ["./mcpsvr"]
No evidence locations were recorded for this file. Raw result