feat: add native HTTP/SSE transport mode (--transport flag)

[leonaburime-ucla]

Problem

codebase-memory-mcp only supports stdio transport. During long or idle agent sessions the stdio pipe silently dies — the server process stays alive but every tool call returns connection closed: EOF. Only a full client restart fixes it. Spawning new subagents doesn't help since all agents share the same CLI-level stdio pipe.

Workaround (painful)

Wrapping with mcp-proxy as a persistent HTTP/SSE daemon works, but requires a separate LaunchAgent and manual config changes.

Proposed fix

Add a --transport sse --port N flag so the server can run as a persistent daemon:

codebase-memory-mcp --transport sse --port 9751

This pattern is already used by Playwright MCP, filesystem MCP, and others. In most MCP server frameworks it's ~10 lines of change.

Benefits

• Survives idle sessions without losing state
• Works across subagents (each gets its own HTTP connection, not a shared pipe)
• Auto-reconnects without server restart

Environment

• codebase-memory-mcp 0.8.1, macOS 15
• Reproduced: every session idle >~20 min causes silent stdio pipe death

[DeusData]

Thanks for the request and for describing the failure mode. I’ve triaged this under MCP-client / transport integration.

Native HTTP/SSE transport is a valid architectural request for long-lived sessions and client reconnects. It is not a short-term drop-in change for this C server, because the transport has lifecycle and local-listener semantics that need to be designed deliberately. For now, mcp-proxy remains the practical workaround; this issue gives us the concrete use-case to revisit native transport support.