Skip to main content
Glama
deenesjakoruzh
baochaofan0404

gpt-dsh-bridge

by baochaofan0404
OverviewSchemaSourceRelated ServersDiscussionsAdmin
TypeScript
Local

GPT↔DSH Bridge

A bridge for ChatGPT web to drive DeepSeek Harness (DSH) for automated task execution.

GPT is the commander, DSH is the executor. An MCP protocol links them together.

English · 中文 | MIT License

Quick Start

git clone https://github.com/baochaofan0404/gpt-dsh-bridge.git
cd gpt-dsh-bridge
pnpm install
pnpm build
pnpm start

Then connect to the exposed https://<tunnel>/mcp endpoint via ChatGPT Developer Mode (Connectors → Add MCP), and you can use tools like dsh_task_submit in the conversation to drive DSH to execute tasks.

Related MCP server: dsh-chatgpt-bridge

Architecture

┌─────────────────┐     HTTPS MCP       ┌──────────────────────────┐     ACP      ┌──────────┐
│  ChatGPT Web    │ ─────────────────►  │  Bridge Orchestrator      │ ───────────► │  DSH     │
│  (MCP Host)     │                     │  (Node · Windows/WSL)     │   stdio RPC  │ (WSL)    │
└─────────────────┘                     │                           │              └──────────┘
     ▲                                   │  · 13 MCP 工具            │
     │ 结果回投（模型自然写入对话）         │  · 任务状态机             │
     │                                   │  · conversation↔session   │
     └───────────────────────────────────┘    映射 + 持久化           │

Core Design Principles:

  • DSH stays untouched: reuse the existing ACP server, zero intrusion

  • Bridge is a thin layer: only task orchestration, state management, and protocol adaptation

  • ChatGPT side connects via native Developer Mode MCP (channel verified at E:\mcpgpt)

MCP Tool Contract (currently 13)

工具

用途

dsh_task_submit

Submit a new task, returns taskId

dsh_task_status

Query task status

dsh_task_result

Get the final task result

dsh_task_cancel

Cancel a running task

dsh_task_revise

Revise based on current results and re-execute

dsh_task_approve

Manually approve HUMAN_REQUIRED tasks

dsh_task_review

Manually review task results

dsh_task_list

List all tasks under a project

dsh_project_create

Create/switch workspace project

dsh_project_status

View project status

dsh_session_list

List DSH sessions

dsh_session_archive

Archive completed sessions

dsh_session_rotate

Rotate DSH session (when context is too long)

Task State Machine

CREATED → SUBMITTED → RUNNING → SUCCESS → COMPLETED
                           ↘ FAILED
                           ↘ TIMEOUT
                           ↘ CRASHED
                           ↘ CANCELLED
                           ↘ HUMAN_REQUIRED → (approve) → RUNNING
                                            → (reject) → CANCELLED

See designs/state-machine.md for the complete state machine definition.

Directory Structure

gpt-dsh-bridge/
├── src/
│   ├── mcp/           # MCP server + 13 个工具实现
│   ├── task/          # TaskManager + TaskStore（SQLite）
│   ├── workspace/     # 项目/工作区管理
│   ├── errors.ts      # BridgeError + error codes
│   └── config.ts      # 配置加载
├── designs/           # 架构设计文档
├── docs/              # 用户文档
├── scripts/           # 启动/隧道脚本
└── recon/             # 侦察报告（原始研究材料）

Dependencies

  • Node.js ≥ 22

  • DeepSeek Harness (DSH) installed and running (pnpm dsh --profile web)

  • OpenAI MCP Tunnel or cloudflared (HTTPS tunnel)

  • ChatGPT Developer Mode (Pro/Plus/Enterprise)

License

MIT © 2026 baochaofan0404

This server cannot be deployed

One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

Maintenance

ActivityMaintained
ResponsivenessNo issues

Resources

Related MCP Connectors

Related MCP Servers