AI Agent Hub
Back to plugins
🧰

dsh-plugin-mcp

Web Tools Updated 2026.09.09

Run the following command in DeepSeek Harness:

dsh plugin install menotbobbybrown/dsh-plugin-mcp

Paste the following prompt into your AI chat to install this plugin:

Install the plugin in DeepSeek Harness by running dsh plugin install menotbobbybrown/dsh-plugin-mcp; the source code is available at https://github.com/menotbobbybrown/dsh-plugin-mcp

About this plugin

You have likely already stood up MCP servers for GitHub, PostgreSQL, the filesystem, and beyond, yet the moment you open DeepSeek Harness to write agents, those tools sit scattered across separate processes with no automatic registration and no unified lifecycle management. dsh-plugin-mcp fills that gap. It is a universal MCP bridge for DSH that attaches any MCP server over stdio, SSE, or WebSocket with zero glue code.

Core capabilities span three layers. On the tool side, MCP tools are dynamically registered into the DSH tool registry with JSON Schema validation, and every tool name is automatically prefixed with its server identity (e.g. github_create_issue, postgres_query) to eliminate cross-server collisions. On the resource and prompt side, MCP resources and prompt templates are exposed directly into the agent context, removing the need to hand-wire them together. On the runtime side, a scale engine built for tens of thousands of servers uses a dormant-by-default process pool with JIT worker activation and an LRU idle-process reaper that persists structured previews as local artifacts; each server runs in an isolated lifecycle with its own timeout and non-blocking error handling, all constructed on native Cordis lifecycle hooks.

It is a natural fit for developers and engineering teams who have already invested in the MCP ecosystem and want a Claude Code-level tool-integration experience without leaving the DSH environment.

Use Cases

  • Batch-attach existing MCP servers (GitHub, PostgreSQL, filesystem, etc.) to DSH agents without hand-writing adapters.
  • Scale to large MCP server pools with on-demand JIT worker activation and automatic idle-process reaping to control resource costs.
  • Expose tools, resources, and prompt templates across multiple MCP servers with automatic name-prefixing to avoid collisions.

Best For

  • Developers who have already invested in the MCP ecosystem and want a near Claude Code-level tool-integration experience in DSH.
  • Engineering teams that need unified process and lifecycle management across dozens to tens of thousands of MCP servers.
  • Developers who want MCP resources and prompt templates injected directly into agent context without manual wiring.