AI Agent Hub
Back to plugins
dsh-llm-cursor preview

dsh-llm-cursor

Client Updated 2026.08.26

Run the following command in DeepSeek Harness:

dsh plugin install NOirBRight/dsh-llm-cursor

Paste the following prompt into your AI chat to install this plugin:

Run dsh plugin install NOirBRight/dsh-llm-cursor in your DeepSeek Harness terminal; the plugin source is hosted at https://github.com/NOirBRight/dsh-llm-cursor.

About this plugin

If you already hold a Cursor subscription and want to route those models through your own DeepSeek Harness agent loop, dsh-llm-cursor was built for that. It authenticates via Cursor's private Deep Control PKCE endpoint and sends chat requests over HTTP/2 + protobuf to the AgentService/Run interface, letting DSH remain the sole agent loop and tool executor without requiring a separate API plan.

Core capabilities include browser-based PKCE login with sessions stored locally on the Host, automatic model catalog discovery with thinking-level SKU collapsing, flexible Fast and 1M context-tier management, live subscription usage display, and a tiered retry policy with full Run lifecycle and binding management. DSH keeps all tool-calling, MCP integration, and multi-model orchestration; Cursor serves purely as a model route.

A critical caveat: this plugin talks to Cursor's unofficial private client endpoints. Cursor staff have stated that this class of usage violates their Terms of Service section 1.5, and accounts may be restricted or permanently banned. It is intended for experienced DSH users who already hold a Cursor subscription and have fully weighed that risk; it is not suitable for teams that treat a Cursor account as production-critical infrastructure.

Screenshots

Use Cases

  • Route Cursor subscription models through the DSH agent loop for chat
  • Orchestrate multi-model and tool calls under a single DSH agent loop
  • Manage Cursor PKCE sessions locally, fetch model catalogs, and organize context tiers

Best For

  • DSH power users with an existing Cursor subscription
  • Independent developers orchestrating multi-provider models on one platform
  • Experienced workflow users who understand the ban risk of unofficial endpoints