Gemini CLI Bridge

OpenAI-compatible API bridge for Google's Gemini CLI. Lets any OpenAI-compatible client (OpenCode, Open WebUI, Cursor, etc.) use Gemini models — including the full CodeAssist API with Google OAuth.

Built on @Intelligent-Internet/gemini-cli-mcp-openai-bridge and the Gemini CLI.

---

What This Fork Fixes

The upstream bridge had a bug where tool schemas containing $ref/$defs (used by AI coding agents like OpenCode) were rejected by the Gemini CodeAssist API with:

Schema.ref 'QuestionPrompt' was set alongside unsupported fields.

Root cause: sanitizeGeminiSchema() stripped $ref but left $defs/definitions intact and never resolved $ref references inline. The Gemini API rejects schemas containing these JSON Schema keywords.

Fixes applied:

• Added resolveRef() to inline $ref/ref references from $defs/definitions before sanitization
• Added ref, $defs, definitions to the unsupported keys list
• Added this.config.setModel() so model selection in clients actually switches the Gemini model
• Added Geminie 3.x models (gemini-3-flash-preview, gemini-3-pro-preview, gemini-3.1-pro-preview, gemini-3.1-flash-lite)
• Handles both JSON Schema $ref and Google protobuf ref formats

---

Quick Start

npm install -g @intelligentinternet/gemini-cli-mcp-openai-bridge

# Start with edit mode (allows file writing)
gemini-cli-bridge --mode edit --port 9000

Bridge starts at http://127.0.0.1:9000/v1 — works with any OpenAI-compatible client.

---

Models

The bridge exposes these models via /v1/models:

Model ID	Gemini API Model
gemini-3.1-pro-preview	Gemini 3.1 Pro
gemini-3.1-flash-lite	Gemini 3.1 Flash Lite
gemini-3-pro-preview	Gemini 3 Pro
gemini-3-flash-preview	Gemini 3 Flash
gemini-2.5-pro	Gemini 2.5 Pro
gemini-2.5-flash	Gemini 2.5 Flash

---

OpenCode Integration

Add this provider to ~/.config/opencode/opencode.json:

{
  "provider": {
    "gemini-bridge": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Gemini CLI Bridge",
      "options": {
        "baseURL": "http://127.0.0.1:9000/v1"
      },
      "models": {
        "gemini-3.1-pro-preview": { "name": "Gemini 3.1 Pro" },
        "gemini-3.1-flash-lite": { "name": "Gemini 3.1 Flash Lite" },
        "gemini-3-pro-preview": { "name": "Gemini 3 Pro" },
        "gemini-3-flash-preview": { "name": "Gemini 3 Flash" },
        "gemini-2.5-pro": { "name": "Gemini 2.5 Pro" },
        "gemini-2.5-flash": { "name": "Gemini 2.5 Flash" }
      }
    }
  }
}

# Start bridge
GEMINI_MCP_PORT=9000 gemini-cli-bridge --mode edit

# OpenCode sees the models automatically
opencode models gemini-bridge

---

Usage

# Start with defaults
gemini-cli-bridge

# Custom port
GEMINI_MCP_PORT=9000 gemini-cli-bridge

# Enable file writing
gemini-cli-bridge --mode edit --target-dir /path/to/project

# Debug logging
gemini-cli-bridge --debug

Flag	Default	Description
--port / GEMINI_MCP_PORT	8765	Listen port
--mode	read-only	Security mode: read-only, edit, configured, yolo
--target-dir	cwd	Root for file operations
--debug	false	Verbose logging

Authentication

No extra auth config needed. The bridge uses the same cached OAuth credentials as the gemini-cli tool. If gemini-cli can talk to Gemini, so can the bridge.

---

Endpoints

Endpoint	Method	Description
/v1/chat/completions	POST	OpenAI-compatible chat (stream + non-stream)
/v1/models	GET	List available models
/mcp	SSE	MCP protocol endpoint

---

Credits

• Original project: @Intelligent-Internet/gemini-cli-mcp-openai-bridge
• Gemini CLI: google-gemini/gemini-cli
• $ref fix + model switching: bulgadev + OpenCode

---

License

Apache License 2.0 — see LICENSE.