dsh-mcphub

Managing MCP servers inside DeepSeek Harness used to mean hand-editing cordis.patch.yml—typing out command, args, env—then wondering whether things actually connected. dsh-mcphub folds that into a dedicated MCP section of the settings panel: a green dot means the server's tools are registered in the live session and callable right now; a grey dot means the config exists but no tools were registered. Transport type, target address, tool count, and sample tool names are all visible at a glance, so you stop guessing from log files.

The real time-saver is the upgrade and connectivity workflow. For pip-installed stdio servers (scrapling, for example), the plugin compares against the latest PyPI release and runs a one-click pip install --upgrade, handling Windows file locks along the way—stopping holder processes, parking the old .exe, and restoring on failure. npx-based servers are matched against the actual local cache with a "refresh cache" action, and @latest specs are never nagged since they self-update. Adding a new server is a simple form that appends the correct entry to the chosen profile's patch.yml; a DSH restart is all it takes. HTTP servers receive a real MCP initialize handshake that returns serverInfo and latency, while stdio servers get at least an executable-existence check—far better than configuring and hoping.

If your daily workflow involves multiple MCP servers across Windows, macOS, and Linux, this plugin turns scattered YAML edits into a single dashboard. Secrets in headers and env stay in the host process for probing only; the browser-side panel never sees the values, only the key names—so you can keep sensitive credentials off your developer tools.