SeekFleet
Cross-platform control plane for DeepSeek Harness agent fleets with MCP, durable sessions, adaptive routing, policy gates, token budgets, and a phone-friendly LAN dashboard.
README
<div align="center">
<img src="./docs/og.png" alt="SeekFleet — Run agents like a fleet" width="100%" />
SeekFleet
Run agents like a fleet.
Cross-platform control plane for DeepSeek Harness agent fleets — with MCP, durable sessions, adaptive routing, policy gates, token budgets, and a phone-friendly LAN dashboard.
<p> <a href="https://cndoin.github.io/seekfleet/">Live product page</a> · <a href="./INSTALL.md">AI installation contract</a> · <a href="./SKILL.md">Agent Skill</a> · <a href="./SECURITY.md">Security</a> </p>
<p> <a href="https://github.com/cndoin/seekfleet/actions/workflows/ci.yml"><img src="https://github.com/cndoin/seekfleet/actions/workflows/ci.yml/badge.svg" alt="CI" /></a> <a href="https://github.com/cndoin/seekfleet/blob/main/LICENSE"><img src="https://img.shields.io/github/license/cndoin/seekfleet?color=42dce8" alt="MIT License" /></a> <a href="https://nodejs.org/"><img src="https://img.shields.io/badge/node-%3E%3D20-43e0a2" alt="Node.js 20 or newer" /></a> <a href="https://github.com/cndoin/seekfleet/commits/main"><img src="https://img.shields.io/github/last-commit/cndoin/seekfleet?color=648bff" alt="Last commit" /></a> </p>
</div>
Give an AI the repository URL. It can install the Skill, configure MCP, verify the runtime, and explain exactly what changed.
International overview
The product page includes a persistent language switcher with five localized experiences: 简体中文, English, 日本語, 한국어, and Español. The page detects the browser language automatically, accepts a direct ?lang= link, and remembers the visitor's choice.
| Language | Product summary |
|---|---|
| 简体中文 | 面向 DeepSeek Harness Agent 集群的跨平台控制平面,统一调度、监控、预算、策略与局域网控制台。 |
| English | A cross-platform control plane for routing, observing, budgeting, governing, and stopping DeepSeek Harness agent fleets. |
| 日本語 | DeepSeek Harness Agent艦隊をルーティング、監視、予算管理、制御するクロスプラットフォーム基盤。 |
| 한국어 | DeepSeek Harness 에이전트 함대를 라우팅, 관찰, 예산 관리, 정책 적용, 중지하는 크로스 플랫폼 제어 플레인. |
| Español | Un plano de control multiplataforma para enrutar, observar, presupuestar, gobernar y detener flotas de agentes DeepSeek Harness. |
The 30-second version
SeekFleet turns DeepSeek Harness from a collection of processes into an observable, governable fleet:
| Surface | What it gives you |
|---|---|
| MCP server | 20 structured tools with stable envelopes and annotations |
| SDK | One TypeScript API for one-shot tasks, sessions, clusters, and DAGs |
| Control plane | Routing, cache, circuit breakers, budgets, policies, and cancellation |
| LAN dashboard | Live agents, requests, tokens, cost, failures, sessions, and protected controls |
| Agent Skill | One deterministic install path for Codex, Claude, Cursor, Gemini, and .agents clients |
Give this repository to an AI
Copy this prompt as-is:
Install SeekFleet from https://github.com/cndoin/seekfleet.
Read INSTALL.md first, install it for this AI client, configure its MCP server,
and verify the installation without exposing credentials.
The machine-readable contract lives in INSTALL.md. It includes Windows PowerShell, Linux, and macOS paths, user/project Skill scopes, verification, update, and rollback behavior.
Install from source
Requirements: Node.js 20+, npm 10+, and @deepseek-ai/dsh or an explicit DSH_MODULE_ROOT.
git clone https://github.com/cndoin/seekfleet seekfleet
cd seekfleet
npm ci
npm run build
node dist/bin/seekfleet.js skill install --target auto
node dist/bin/seekfleet.js inspect
Install the Skill into every supported AI client:
node dist/bin/seekfleet.js skill install --target all --force
For a project-local Skill, use --scope project; it installs into .agents/skills/seekfleet without changing the user profile.
Start the fleet
node dist/bin/seekfleet.js serve-mcp
Run MCP and the phone dashboard together on a trusted LAN:
node dist/bin/seekfleet.js serve-mcp \
--dashboard \
--dashboard-host 0.0.0.0 \
--dashboard-port 8787
Open the printed URL on a phone connected to the same network. The dashboard is bearer-token protected; do not expose it directly to the public internet.
SDK at a glance
import { SeekFleet } from "seekfleet";
const fleet = new SeekFleet();
const review = await fleet.run("Review this module for race conditions");
console.log(review.answer);
const clusterId = fleet.cluster({
routing: "adaptive",
instances: [
{ label: "code-a", tags: ["code"] },
{ label: "code-b", tags: ["code"] },
{ label: "review", tags: ["review"] },
],
});
const result = await fleet.clusterRoute(clusterId, {
task: "Audit the authentication flow",
tags: ["review"],
timeoutMs: 120_000,
});
await fleet.clusterShutdown(clusterId);
DshPlugin remains available as a deprecated source-compatibility alias. MCP tool names retain the dsh_* prefix so existing integrations keep working.
Choose your execution mode
| Mode | Best for | Control surface |
|---|---|---|
| One-shot | One bounded task | dsh_run |
| Session | Long-running, cancellable work | dsh_session_* |
| Cluster | Parallel or repeated tasks | dsh_cluster_* |
| DAG | Dependent task graphs | dsh_dag_run |
Routing strategies include least-loaded, round-robin, tag, adaptive, and random.
Architecture
AI client (Codex / Claude / Cursor / Gemini / Hermes / OpenClaw)
│
MCP or SDK
│
SeekFleet
┌───────────────────┼───────────────────┐
│ │ │
Durable sessions Agent clusters Policy gate
│ routing · cache · DAG │
└───────────────────┼───────────────────┘
│
DeepSeek Harness
│
tools · models · subprocesses
Cross-platform configuration
PowerShell:
$env:DSH_MODULE_ROOT = "C:\Tools\deepseek-harness"
$env:DSH_HOME = "$env:USERPROFILE\.dsh"
npm run serve-lan
Bash or zsh:
export DSH_MODULE_ROOT=/opt/deepseek-harness
export DSH_HOME="$HOME/.dsh"
npm run serve-lan
Use OS-native absolute paths. Do not copy Windows drive paths into Linux or macOS configuration.
Environment variables
| Variable | Default | Purpose |
|---|---|---|
DSH_MODULE_ROOT |
auto-detected | Installed DeepSeek Harness package |
DSH_HOME |
~/.dsh |
Profiles and persistent runtime state |
CODEX_HOME |
~/.codex |
Codex configuration root |
SEEKFLEET_DASHBOARD |
0 |
Start the dashboard with MCP when set to 1 |
SEEKFLEET_DASHBOARD_HOST |
127.0.0.1 |
Dashboard bind host |
SEEKFLEET_DASHBOARD_PORT |
8787 |
Dashboard TCP port |
SEEKFLEET_DASHBOARD_TOKEN |
random | Fixed dashboard bearer token |
The previous DSH_DASHBOARD* names remain accepted as compatibility aliases.
Project map
src/— SDK, routing, sessions, policies, metrics, dashboard, and MCP serverbin/seekfleet.ts— cross-platform CLI entry pointSKILL.md— compact AI operating instructionsagents/openai.yaml— AI client discovery metadatadocs/— GitHub Pages product site and social preview assetexamples/— MCP, SDK, adapter, and cluster examplestests/— 94 behavior and integration tests
Development
npm ci
npm run format:check
npm run typecheck
npm run lint
npm test
npm run build
CI runs on Windows, Ubuntu, and macOS with Node.js 20, 22, and 24. The live product page is published from docs/ by GitHub Pages.
Security
SeekFleet executes real agent tools and subprocesses. Configure policies, budgets, finite timeouts, and restricted workspaces before autonomous use. See SECURITY.md for the threat model and disclosure process.
License
MIT — see LICENSE.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。