Guide for wiring existing Model Context Protocol (MCP) servers (filesystem, GitHub, Postgres) into DeepSeek Harness so the same MCP setup works across dsh, Claude Code, and Cursor.
DeepSeek Harness MCP Server Bridge Configuration
Overview
DeepSeek Harness (dsh) added MCP support in its packaged Python runtime, meaning MCP servers you already run for Claude Code or Cursor can potentially be reused with dsh instead of maintaining separate configs per agent tool.
> ⚠️ dsh is a developer preview — MCP wiring syntax may change. Verify current config keys against the DeepSeek Harness docs before deploying.
Shared MCP server pool (reuse across tools)
{
"mcpServers": {
"filesystem": {
"command": "npx",
"args": [
"-y",
"@modelcontextprotocol/server-filesystem",
"/path/to/project-sandbox"
]
},
"github": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-github"],
"env": {
"GITHUB_PERSONAL_ACCESS_TOKEN": "ghp_your_read_scoped_token_here"
}
}
}
}
Bridging checklist
1. Start from one canonical MCP server list — keep the mcpServers block above as your source of truth and point each tool's config (Claude Desktop config, Cursor settings, dsh config) at the same server commands so credentials and scopes stay consistent.
2. Scope tokens per server, not per tool — issue read-only GitHub/DB tokens once, then reference them via environment variables from every agent runtime that connects.
3. Sandbox filesystem access — restrict the filesystem server to a dedicated project folder rather than your whole home directory, regardless of which agent (dsh, Claude, Cursor) is calling it.
4. Re-check on upgrades — since dsh is under active development, re-verify MCP config compatibility after every dsh version bump.
Why add this to SimplyUtils
Teams adopting DeepSeek Harness alongside Claude Code/Cursor don't want three divergent MCP setups — this resource frames the problem as 'one MCP server pool, multiple front-ends' rather than duplicating server configs per tool.