dsh-mcp-manager

Managing MCP servers in the DeepSeek Harness (DSH) Web environment used to mean hand-editing a cordis.patch.yml text file: easy to mangle the YAML, no visual feedback on whether a server actually registered its tools, and every change demanded a full process restart. dsh-mcp-manager moves the entire workflow into the Settings panel. Under Settings → MCP Servers, a straightforward form covers adding, editing, and deleting servers over stdio or streamable-http, with full support for command, args, cwd, env, url, and headers. On save, the host picks up the change via Cordis HMR and re-applies it transactionally, so dsh-mcp-client reconnects in place — no restart, no downtime. Every row carries a live status badge: green with a tool count when connected, grey with hover details when offline, plus a manual refresh button and a 30-second polling fallback.

Under the hood the writer auto-quotes reserved YAML indicators (-y, @pkg/name, etc.) so the generated patch always parses, rewrites only its own mcp-client rows, and leaves the rest of the file untouched. Concurrent edits are serialized through a single queue so nothing interleaves.

If you are running several MCP servers (file system, databases, internal APIs) inside a DSH Web profile and want a visual, zero-restart way to keep them in sync, this plugin is built for you. It introduces no separate store — everything lives in the same profile patch layer — and removes cleanly by dropping the dependency and the single patch row.