dsh-claude-bridge

DeepSeek Harness can already connect to Anthropic via OAuth, but third-party harness usage on that route is billed as extra usage per token rather than against your Claude Pro or Max plan limits. dsh-claude-bridge takes a different path: it drives the already-authenticated Claude Code CLI through the Agent SDK, so inference and tool-call usage lands on the subscription you already pay for.

The plugin supports streaming text and reasoning output, multi-round and parallel tool calls (with precise tool-use id pairing), token accounting including cache reads and writes, finish reasons, model listing and resolution, reasoning effort control, and an optional one-million-token context window. On the safety side, the child process reads no settings files at all, and environment variables such as ANTHROPIC_API_KEY that could redirect billing to a different account are explicitly stripped before the CLI runs.

It is designed for users who hold a Claude Pro or Max subscription and want to route Claude model calls through DeepSeek Harness while staying within plan limits. Keep in mind that conversation history is carried as inline prompts rather than written to the CLI's session file, so cross-turn prefix caching does not apply (it does within a single tool round), which can increase cost on long conversations. Anthropic has also made and then walked back a change to how Agent SDK usage is billed, so it is worth confirming the current policy before relying on this approach.