AI Agent Hub
Back to plugins
🤖

dsh-mcp-adapter

Model Inference Updated 2026.08.22

Run the following command in DeepSeek Harness:

dsh plugin install NexusAgentX/dsh-mcp-adapter

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

Run dsh plugin install NexusAgentX/dsh-mcp-adapter in the DeepSeek Harness web terminal to install the plugin; the repository is available at https://github.com/NexusAgentX/dsh-mcp-adapter. After installation, restart dsh web and hard-refresh the browser to access the /mcp menu.

About this plugin

The default MCP client in DeepSeek Harness connects every registered server at startup and registers each MCP tool schema as a native function, so every request pays the context cost of hundreds of schemas. dsh-mcp-adapter replaces that model with the pi-mcp-adapter contract: a single MCP proxy tool, lazy server startup, and a search-describe-call workflow. A metadata cache ensures search results are available even before a live connection is established.

Configuration lives in the web /mcp menu: one-click add for DeepWiki, Context7, Notion, GitHub, Chrome DevTools, and more, with OAuth browser authorization flows. You can also hand-write .mcp.json to share the same server list across Cursor, Claude Code, and other MCP hosts. Hot-path tools can be promoted to native dsh tools via directTools, mcpScript chains multiple MCP calls in a single JavaScript request, and approveTools inserts a confirmation dialog before sensitive tools execute.

If you mount several MCP servers and want to reserve context window for actual reasoning, or need a single MCP config shared across multiple AI coding hosts, this plugin fits the job. Just avoid pointing it at the same servers as @deepseek-ai/dsh-mcp-client, or both will double-connect and clash on tool names.

Use Cases

  • Running multiple MCP servers and burning context window on schemas
  • Sharing one MCP config across Cursor, Claude Code, and other hosts
  • Searching and calling MCP tools on demand instead of loading everything at startup

Best For

  • DeepSeek Harness developers mounting several MCP servers
  • AI engineers who want to free up context window for reasoning
  • Power users sharing MCP configs across multiple AI coding hosts