dsh-harness-mcp-server
Expose DeepSeek Harness agent capabilities as an MCP server, letting any MCP client (e.g. Hermes) drive Harness to execute real coding tasks.
Hermes is the brain (pro), Harness is the arms (flash) — 1+1>2.
接入你的智能体
從發布者安裝。不是 PlugMCP 端點。
遠端 URL 還是本機指令? 兩種安裝方式的差異
來自發布者
Expose DeepSeek Harness agent capabilities as an MCP server, letting any MCP client (e.g. Hermes) drive Harness to execute real coding tasks.
Hermes is the brain (pro), Harness is the arms (flash) — 1+1>2.
Expose DeepSeek Harness agent capabilities as an MCP server (brain=Hermes, arms=Harness)
Harness ships a powerful agent runtime (tools, LLM, agents, sessions), but it is a Cordis app, not something another agent can call. This plugin turns Harness inside-out: it starts a real MCP server (StreamableHTTP) inside Harness and bridges Harness's core services — ctx.agents, ctx.agentPresets, ctx.tools — so an external "brain" can delegate real work to Harness's "arms".
Hermes (MCP client, brain)
│ agent_run / task_inbox (HTTP)
▼
dsh-harness-mcp-server (MCP server, :8090)
│ ctx.agents.create → mount 'standard' preset
▼
Harness agent (flash) — full toolset: bash, fs, todo, web…
Every task result is structured:
{
"sessionId": "...",
"assistantText": "final answer",
"toolCalls": [{ "name": "bash", "args": "..." }],
"toolResults": ["command output"],
"changes": "what was changed",
"verification": "how it was verified",
"leftovers": "open issues"
}
This closes the loop between the client's persistent memory and Harness's coding: memory is fed into each task as context, and the result (changes / verification / leftovers) can be persisted back to the client's memory for the next run.
npm install @chushixixin/dsh-harness-mcp-server
Then reference the plugin from your Harness workspace (see the cordis patch below).
Clone this repo inside your Harness workspace under packages/mcp/harness-mcp-server/ (the pnpm workspace matches packages/*/*, two levels deep):
cd /path/to/deepseek-harness
mkdir -p packages/mcp/harness-mcp-server
# copy this repo's files there, then:
corepack pnpm install
Register it in tsconfig.host.json references and tsconfig.base.json paths (see the Harness plugin docs), then build:
corepack pnpm exec tsc -b packages/mcp/harness-mcp-server
corepack pnpm run build:lib:host
export DEEPSEEK_API_KEY=...
corepack pnpm dsh web --patch ./packages/mcp/harness-mcp-server/cordis.yml
The MCP server listens on 127.0.0.1:8090 (StreamableHTTP). Point any MCP client at http://127.0.0.1:8090/mcp.
⚠️ Security: by default the server binds to
127.0.0.1only. It exposes unauthenticated remote code execution — do not bind it to0.0.0.0or expose it to the internet/LAN without adding authentication, TLS, and a reverse proxy first.
printf 'n\nY\n' | hermes mcp add harness_plugin --url http://127.0.0.1:8090/mcp
- insert:
- id: harness-mcp-server
name: '@chushixixin/dsh-harness-mcp-server'
config:
http: true
port: 8090
host: 127.0.0.1 # 默认仅本机; 暴露前必须加认证
# authToken: 'your-secret-token' # 可选: Bearer token 认证
# workspaceRoots: ['/workspace'] # 可选: cwd 白名单
This is best used as a fallback tool, not a daily driver: for everyday code edits, drive your primary agent directly. Reach for this when you need context isolation (huge refactors that would blow the client's context) or parallel execution of unrelated tasks.
dsh headless).workspace-write): install bubblewrap on the host, or the sandbox will refuse write commands.McpServer + transport (an MCP McpServer connects to a single transport).MIT
Expose DeepSeek Harness agent capabilities as an MCP server (brain=Hermes, arms=Harness)
工具
尚無工具清單。遠端伺服器會在首次健康檢查後取得工具清單。
常見問題
在本頁複製適用於 Cursor、Claude Desktop、Claude Code 或 VS Code 的 stdio 設定。它使用發布者的安裝方式,而不是 PlugMCP 代理伺服器。
這筆記錄未列出必要的環境變數。
Dsh Harness 透過 stdio 在本機執行。
上次檢查時間:2026-10-27。 Expose DeepSeek Harness agent capabilities as an MCP server (brain=Hermes, arms=Harness)