Claudeseek Code — Development skill for Claude Code

claudeseek

**Run full Claude Code on DeepSeek**, against the direct API, no OpenRouter and no middlemen. Tools, skills, hooks, MCP, CLAUDE.md, subagents, and images. The only thing that changes is the model answering.

Claude Code stays Claude Code: it's just told to talk to a local gateway that translates Anthropic's protocol to DeepSeek's, and nothing else.

Not affiliated with Anthropic or DeepSeek. It's a bridge between two public APIs.

---

Index

• What it is — technical spec
• What it's for
• How it works
• Replicating it in your Claude Code
• Usage
• Models
• Images
• Security
• Limitations
• Adding a model
• Troubleshooting
• Repo structure
• Status
• Uninstalling
• License

---

What it is

A **local HTTP gateway** that implements Anthropic's Messages API and translates it to DeepSeek's OpenAI-compatible API, in both directions. It sits between a Claude Code process and `api.deepseek.com`.

Spec

Listens on	127.0.0.1:4319 (configurable in config.json), loopback only
Speaks	Anthropic Messages API ↔ DeepSeek (/chat/completions)
Upstream	https://api.deepseek.com (configurable)
Runtime	Node.js 18+, no external dependencies (uses native fetch)
Routes implemented	POST /v1/messages, POST /v1/messages/count_tokens, GET /v1/models, GET /v1/models/:id, GET /health
Any other route	untouched passthrough to api.anthropic.com
Configuration	gateway/config.json (read once at startup)
Secrets	~/.claude/deepseek-gateway/.env (never committed)

What it translates

Capability	Status
Text, SSE streaming and non-streaming	Yes
Tools (tool_use / tool_result)	Yes