dsh-scan-mcp

Windows developers who juggle Claude Code, Codex, and CodeBuddy simultaneously often end up with a dozen MCP servers scattered across three home-directory config files. Which one is actually reachable, which is 404-ing because of a typo'd package name, which stdio process never answers the initialize handshake—answering those questions used to mean running a probe command for each entry by hand. dsh-scan-mcp consolidates that workflow: it auto-discovers all three config files, deduplicates across agents, then performs a genuine MCP initialize handshake on every enabled server (spawning the real process and reading the JSON-RPC response over stdio, or POSTing and parsing both JSON and SSE responses over streamable-http). Online and offline status, latency, and the concrete failure reason all show up in one glance.

The control panel offers per-server sliding toggles (state persisted to ~/.dsh/dsh-scan-mcp.json across restarts), a Reconnect button for a single server without a full rescan, and a hard rule that disabled servers are never probed. Type /mcp in any session to pop up the panel, or open the card-layout view under Settings. The model can also call the mcp_discovered_catalog tool to fetch the full inventory. On the security side the plugin surfaces only environment-variable and header names—credential values are never read or displayed—and each probe terminates its process the moment the handshake completes.

Built for Windows users running dsh web who maintain multi-agent MCP configurations and want a single dashboard to audit connection health, spot 404 / timeout / corrupted-cache issues at a glance, and flip servers on or off without touching any config file.