dsh-claude-cli

dsh-claude-cli bridges the gap between DeepSeek Harness and Claude Code CLI, enabling agent frameworks to leverage locally installed Claude environments directly. Many developers already use the Claude Code CLI interactively but seek to integrate its model capabilities into more powerful agent systems; the plugin runs the CLI as a subprocess without requiring an API key, reusing existing login states and subscriptions. This eliminates the hassle of key management while maintaining harness control as the core agent, automatically disabling CLI-specific features like autonomous tool loops and memory files.

Its core strength lies in intelligent integration and transparent bridging: the plugin converts harness tool definitions into an MCP server, allowing Claude models to generate structured tool calls while execution remains under harness authority. Configuration is flexible, supporting custom executable paths, working directories, and timeout settings, plus extra arguments for CLI tuning. Users find familiar Claude model aliases (e.g., fable, opus) in the model picker and experience streamed responses seamlessly.

The plugin is ideal for individual developers or small teams already using the Claude Code CLI, who wish to incorporate Claude models into DeepSeek Harness agent workflows for automation, code generation, or complex reasoning. It suits those needing model inference without managing separate API keys, but users must note Anthropic's consumer terms may restrict programmatic access. Evaluate your plan's usage terms before adoption, and prioritize this plugin for scenarios that would otherwise involve manual Claude Code work.