PulseAugur
EN
LIVE 04:30:06

MCP session bug fixed by separating protocol and application state

A technical article details a reproducible issue with an MCP session across different replicas, leading to a 400 error after an initial 200 success. The problem arises when a proxy sends a tool call to a different replica than the one that initially established the session. The author proposes a solution involving separating protocol state from application state, ensuring that protocol facts are available on the current request and durable application state uses explicit, shared domain handles. This approach aims to make the application contract inspectable and support stateless replicas, moving away from sticky sessions and shared opaque session stores. AI

RANK_REASON Technical article detailing a specific bug and proposed solution in a protocol. [lever_c_demoted from research: ic=1 ai=0.4]

Read on dev.to — MCP tag →

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →

MCP session bug fixed by separating protocol and application state

COVERAGE [1]

  1. dev.to — MCP tag TIER_1 English(EN) · xn ·

    MCP initialize Removed: Reproducing Unknown MCP Session Across Replicas

    <p>The first MCP request returns 200. The next <code>tools/call</code> returns 400 with <code>Unknown MCP session</code>. The endpoint did not change, the tool schema is valid, and the network is healthy. What changed was the replica.</p> <p>In a legacy Streamable HTTP flow, <cod…