dsh-chatgpt-bridge

For people who already run DeepSeek Harness (DSH) locally, dsh-chatgpt-bridge solves a specific connection problem: how to make ChatGPT Web a remote command and observation surface for DSH sessions without changing DSH core or its security model. It links ChatGPT Web, the bridge, and local DSH through the Model Context Protocol, with control flowing from ChatGPT Web toward DSH; it does not make DSH call ChatGPT, nor does it route DSH model requests through ChatGPT.

The plugin focuses on session supervision rather than simple chat forwarding. It can create, view, continue, and supervise DSH agent sessions, and manage goal-oriented work through the Goal Control Plane, while DSH still keeps its own session log, agent loop, tools, skills, subagents, workflows, approvals, sandbox, and workspace security model. For tasks that actually need to do something, the bridge also exposes write-capable and action-capable MCP tools, making it useful as a friendlier remote console for ChatGPT Web instead of a read-only status viewer.

It is best suited to advanced users who already run a local or self-hosted DSH, are comfortable with DSH workflows, and take MCP security boundaries seriously. Because the bridge is local-first and requires ChatGPT Web to connect through an OpenAI-supported secure MCP tunnel, token handling, workspace registration, approval policy, and runtime lifecycle all matter; it is strongest for personal setups, debugging, and supervised automation where DSH remains the execution and security center.