opencode-windsurf-auth

Opencode plugin for Windsurf/Codeium authentication - use Windsurf models in Opencode.

Features

• OpenAI-compatible /v1/chat/completions interface with streaming SSE
• Automatic credential discovery (CSRF token, port, API key)
• Transparent REST↔gRPC translation over HTTP/2
• Zero extra auth prompts when Windsurf is running
• Opencode tool-calling compatible: tools are planned via Windsurf inference but executed by Opencode (MCP/tool registry remains authoritative)

Overview

This plugin enables Opencode users to access Windsurf/Codeium models by leveraging their existing Windsurf installation. It communicates directly with the local Windsurf language server via gRPC—no network traffic capture or OAuth flows required.

Prerequisites

1. Windsurf IDE installed - Download from windsurf.com
2. Windsurf running - The plugin communicates with the local language server
3. Logged into Windsurf - Provides API key in ~/.codeium/config.json
4. Active Windsurf subscription - Model access depends on your plan

Installation

bun add opencode-windsurf-auth@beta

Opencode Configuration

Add the following to your Opencode config (typically ~/.config/opencode/config.json). The plugin starts a local proxy server on port 42100 (falls back to a random free port and updates chat.params automatically). The full model list with variants is in opencode_config_example.json; thinking vs non-thinking are separate models, while variants are only for performance tiers (low/high/xhigh/etc.).

{
  "$schema": "https://opencode.ai/config.json",
  "plugin": ["opencode-windsurf-auth@beta"],
  "provider": {
    "windsurf": {
      "npm": "@ai-sdk/openai-compatible",
      "options": {
        "baseURL": "http://127.0.0.1:42100/v1"
      },
      "models": {
        "claude-4.5-opus-thinking": {
          "name": "Claude 4.5 Opus Thinking (Windsurf)",
          "limit": {
            "context": 200000,
            "output": 8192
          }
        },
        "gpt-5.1-codex-max": {
          "name": "GPT 5.1 Codex Max (Windsurf)",
          "limit": {
            "context": 200000,
            "output": 8192
          },
          "variants": {
            "low": {},
            "medium": {},
            "high": {}
          }
        },
        "gemini-3.0-pro": {
          "name": "Gemini 3.0 Pro (Windsurf)",
          "limit": {
            "context": 200000,
            "output": 8192
          },
          "variants": {
            "minimal": {},
            "low": {},
            "medium": {},
            "high": {}
          }
        },
        "minimax-m2.1": {
          "name": "Minimax M2.1 (Windsurf)",
          "limit": {
            "context": 200000,
            "output": 8192
          }
        },
        "glm-4.7": {
          "name": "GLM 4.7 (Windsurf)",
          "limit": {
            "context": 200000,
            "output": 8192
          }
        },
        "glm-4.7-fast": {
          "name": "GLM 4.7 Fast (Windsurf)",
          "limit": {
            "context": 200000,
            "output": 8192
          }
        }
      }
    }
  }
}

After saving the config:

opencode models list                                            # confirm models appear under windsurf/
opencode chat --model=windsurf/claude-4.5-opus "Hello"          # quick smoke test

Keep Windsurf running and signed in—credentials are fetched live from the IDE process.

Project Layout

src/
├── plugin.ts              # Fetch interceptor that routes to Windsurf
├── constants.ts           # gRPC service metadata
└── plugin/
    ├── auth.ts            # Credential discovery
    ├── grpc-client.ts     # Streaming chat bridge
    ├── models.ts          # Model lookup tables
    └── types.ts           # Shared enums/types

How It Works

1. Credential Discovery: Extracts CSRF token and port from the running language_server_macos process
2. API Key: Reads from ~/.codeium/config.json
3. gRPC Communication: Sends requests to localhost:{port} using HTTP/2 gRPC protocol
4. Response Transformation: Converts gRPC responses to OpenAI-compatible SSE format (assistant/tool turns are not replayed back to Windsurf)
5. Model Naming: Sends both model enum and chat_model_name for fidelity with Windsurf’s expectations
6. Tool Planning: When tools are provided, we build a tool-calling prompt (with system messages) and ask Windsurf to produce tool_calls/final text. Tool execution and MCP tool registry stay on OpenCode’s side.

Supported Models (canonical names)

Claude: claude-3-opus, claude-3-sonnet, claude-3-haiku, claude-3.5-sonnet, claude-3.5-haiku, claude-3.7-sonnet, claude-3.7-sonnet-thinking, claude-4-opus, claude-4-opus-thinking, claude-4-sonnet, claude-4-sonnet-thinking, claude-4.1-opus, claude-4.1-opus-thinking, claude-4.5-sonnet, claude-4.5-sonnet-thinking, claude-4.5-opus, claude-4.5-opus-thinking, claude-code.

OpenAI GPT: gpt-4, gpt-4-turbo, gpt-4o, gpt-4o-mini, gpt-4.1, gpt-4.1-mini, gpt-4.1-nano, gpt-5, gpt-5-nano, gpt-5-codex, gpt-5.1-codex-mini, gpt-5.1-codex, gpt-5.1-codex-max, gpt-5.2 (variants low/medium/high/xhigh + priority tiers). Non-thinking vs thinking are separate model IDs, not variants.

OpenAI O-series: o3, o3-mini, o3-low, o3-high, o3-pro, o3-pro-low, o3-pro-high, o4-mini, o4-mini-low, o4-mini-high.

Gemini: gemini-2.0-flash, gemini-2.5-pro, gemini-2.5-flash, gemini-2.5-flash-thinking, gemini-2.5-flash-lite, gemini-3.0-pro (variants: minimal, low, medium, high), gemini-3.0-flash (variants: minimal, low, medium, high). Thinking versions of Gemini 2.5 are separate models.

DeepSeek: deepseek-v3, deepseek-v3-2, deepseek-r1, deepseek-r1-fast, deepseek-r1-slow.

Llama: llama-3.1-8b, llama-3.1-70b, llama-3.1-405b, llama-3.3-70b, llama-3.3-70b-r1.

Qwen: qwen-2.5-7b, qwen-2.5-32b, qwen-2.5-72b, qwen-2.5-32b-r1, qwen-3-235b, qwen-3-coder-480b, qwen-3-coder-480b-fast.

Grok (xAI): grok-2, grok-3, grok-3-mini, grok-code-fast.

Specialty & Proprietary: mistral-7b, kimi-k2, kimi-k2-thinking, glm-4.5, glm-4.5-fast, glm-4.6, glm-4.6-fast, glm-4.7, glm-4.7-fast, minimax-m2, minimax-m2.1, swe-1.5, swe-1.5-thinking, swe-1.5-slow.

Aliases (e.g., gpt-5.2-low-priority) are also accepted. Variants live under provider.windsurf.models[model].variants; thinking/non-thinking are distinct models.

Development

# Install dependencies
bun install

# Build
bun run build

# Type check
bun run typecheck

# Run tests
bun test

Known Limitations

• Windsurf must be running - The plugin communicates with the local language server
• macOS focus - Linux/Windows paths need verification

Further Reading

• docs/WINDSURF_API_SPEC.md – gRPC endpoints & protobuf notes
• docs/REVERSE_ENGINEERING.md – credential discovery + tooling
• opencode-antigravity-auth – related project

License

MIT