local-llm-proxy

A local API proxy that delegates to the claude CLI. Use your Claude Max subscription in any AI-aware IDE — no separate API key required.

Speaks both Anthropic (/v1/messages) and OpenAI (/v1/chat/completions) formats, so it works with virtually any IDE or tool that supports custom AI providers.

How it works

IDE  →  POST http://127.0.0.1:9099/v1/chat/completions  (OpenAI format)
   or   POST http://127.0.0.1:9099/v1/messages           (Anthropic format)
                     ↓
          parse messages[] + system prompt
                     ↓
     spawn: claude --print --dangerously-skip-permissions
                     ↓
     stream stdout → SSE events (matching request format)
                     ↓
                IDE receives response

The proxy never touches your OAuth token directly — it delegates to the official claude CLI, which manages its own auth. All requests stay on 127.0.0.1.

Requirements

• Node.js 20+
• claude CLI installed and authenticated (claude /login)

Install

npm install -g local-llm-proxy

Or from source:

git clone https://github.com/vxcozy/local-llm-proxy
cd local-llm-proxy
npm install
npm run build

Usage

# If installed globally:
local-llm-proxy

# If running from source:
npm start

# Custom port (default is 9099):
local-llm-proxy 3456

IDE configuration

Zed

Add this to your ~/.config/zed/settings.json (inside the top-level object):

"language_models": {
  "openai": {
    "api_url": "http://127.0.0.1:9099/v1",
    "available_models": [
      {
        "name": "claude-opus-4-6",
        "display_name": "Claude Opus (local proxy)",
        "max_tokens": 16384
      }
    ]
  }
}

When Zed prompts for an OpenAI API key, enter any non-empty string (e.g. x). Then select Claude Opus (local proxy) from the model dropdown in the Assistant panel.

Continue.dev (VS Code / JetBrains)

In ~/.continue/config.json:

{
  "models": [{
    "provider": "anthropic",
    "model": "claude-opus-4-6",
    "apiBase": "http://127.0.0.1:9099",
    "apiKey": "x"
  }]
}

Cursor

Settings > Models > Anthropic > Override Base URL: http://127.0.0.1:9099

Generic (any tool with ANTHROPIC_BASE_URL)

export ANTHROPIC_BASE_URL=http://127.0.0.1:9099

Generic (any tool with OpenAI-compatible endpoint)

Point the base URL to http://127.0.0.1:9099/v1 and use any non-empty API key.

Endpoints

| Method | Path | Format | Description | |--------|------|--------|-------------| | POST | /v1/messages | Anthropic | Messages API (streaming + non-streaming) | | POST | /v1/chat/completions | OpenAI | Chat Completions API (streaming + non-streaming) | | GET | /v1/models | OpenAI | Lists available models | | GET | /health | — | Health check — returns {"ok":true} |

Environment variables

| Variable | Description | |----------|-------------| | CLAUDE_NO_TOOLS | Set to 1 to disable Claude's built-in coding tools (pure chat mode) |

Notes on --dangerously-skip-permissions

The proxy passes this flag so the CLI never blocks waiting for a TTY permission prompt (there is no TTY). Claude's native tools (Bash, file read/write) remain available and are governed by your normal Claude Max permissions.

License

MIT