dsh-tool-explorer

When you have a dozen Skills and several MCP servers wired into dsh, the pain rarely shows up in code — it shows up in configuration. Which skill should have model invocation off? Which MCP server is pointing at the wrong port? How do you carry MCP configs across Claude, Cursor, and Codex without hand-editing a handful of JSON and TOML files? dsh-tool-explorer pulls all of that into a single Web console so you can manage dsh capabilities the way you would manage any back-office panel.

On the Skills side, it merges every source — project, user, bundled, plugin — into one collapsible hierarchy with search, inline frontmatter and Markdown editing, and two independent switches for model-versus-user invocation. Deleting a skill drops it into a recoverable trash bin rather than vanishing it. GitHub installs support batch multi-select, a single tarball fetch per repo, and per-skill conflict reporting. On the MCP side, add, edit, enable, and disable all flow through a hot-patch layer that applies without a restart. A built-in SDK probe runs connection tests in isolation so live instances are never disturbed, and a one-click import scans Claude, Cursor, Codex, Cline, Roo, Continue, and Windsurf config files.

Best for developers who run multiple AI agents under dsh, tweak Skill combinations frequently, or need to reuse MCP server configurations across tools. If you are tired of shelling out to the terminal and hand-flipping through YAML and JSON files, this panel saves real time.