Little Mungbean is a free, open-source, simple, lightweight AI gateway manager. It lets Codex and Claude clients share a local gateway that is easy to switch and monitor. Features include multiple providers, local protocol conversion, model injection, MCP Server and Skills management, usage statistics, and more.
If Little Mungbean helps you, please give us a Star on GitHub. It means a lot to us. Or buy the author a coffee.
You may also become a sponsor. Your info will be shown below or in vendor presets. Email is available at the bottom.
Sponsors
No sponsors yet
Features
- One-click client setup • Install Codex CLI, Codex Desktop, Claude Code, and Claude Desktop with one click.
- Multiple providers • Save multiple provider URLs and API keys, rename and reorder them, and switch between providers quickly.
- Provider switching • After changing a provider URL or API key, select Restart • Connect to restart the selected Agent client and reload its configuration.
- Account mode • Switch between the local gateway and an Agent client’s official account without signing out; the original configuration is restored automatically.
- Independent client connections • Manage installation, runtime, connection status, and gateway routing separately for all four Agent clients.
- MCP Server management and Market • Add MCP Servers locally or from a Market; install, edit, search, enable, disable, and remove them; choose supported Agent clients and open server websites.
- Skills management and Market • Add Skills locally or from a Market; install from a Market or import folders and ZIP files; edit, search, enable, disable, and remove Skills; choose supported Agent clients and open Skill websites. Import Skills into Claude Desktop via Customize > Skills.
- Local protocol conversion • Convert between OpenAI, Anthropic, Gemini, and other protocols. Same-family models are forwarded directly when possible, and conversion can be toggled live. See the FAQ below.
- Gateway model management • Fetch provider model lists and hide outdated or infrequently used models per provider.
- Model list injection • Inject provider models into Agent client model lists so they can be selected directly.
- Provider usage statistics • Track request counts and token usage for each provider.
- Live status detection • Automatically detect whether each Agent client is installed, running, and connected to the gateway.
- Secure configuration • API keys are encrypted on this device, and the gateway listens only on the local loopback address.
- Client connection details • Right-click a client card to open its configuration folder or copy the current local connection’s Base URL and Token for use with other tools.
- Runtime log • Records client startup, connection, restart, model injection, gateway operations, and MCP/Skills changes. The local listening address appears after a successful start or restart; CLI clients also show their configuration and working folders. Copy the connection token from a client card’s context menu.
- Request log • Records the Agent, provider, model, status code, tokens, duration, time to first token, request endpoint, and protocol conversion details.
- Chinese and English interface • The interface and logs are available in Chinese and English.
Client status indicators
- Gray(×) • Not installed
- Gray • Not running
- Blue • Running, not connected to the local gateway
- Green • Connected to the local gateway
Flow indicator
- Black • Not connected
- Green • Connected; the moving dot does not indicate live data transfer
Frequently asked questions
What is local gateway mode?
Agent clients (such as Claude Code and Codex Desktop) connect to Little Mungbean’s local gateway, which forwards requests to the actual AI provider, including official APIs and third-party API relays.
The local gateway handles protocol conversion, request routing, provider switching, and API key management, making it easier for different clients to use different models.
What is account mode?
In account mode, an Agent client connects and chats through your signed-in official AI provider account, without using Little Mungbean’s local gateway routing.
What is protocol conversion?
Protocol conversion automatically translates the API format used by an Agent client into a format supported by an upstream model, converting requests and responses in both directions, including streaming responses.
Local protocol conversion currently supports these formats:
- OpenAI Chat Completions
- OpenAI Responses
- Anthropic Messages
- Gemini GenerateContent
When should I enable local protocol conversion?
Enable local protocol conversion when you want to use models from a different model family, for example:
- Use Claude, Gemini, or other model families in Codex.
- Use GPT, Grok, Gemini, or other model families in Claude.
When should I disable local protocol conversion?
- When you do not need models from other model families.
- Disable local conversion when your provider already converts protocols on its servers to avoid converting twice.
- Grok natively supports OpenAI Responses, so local protocol conversion is usually unnecessary when using Grok with Codex.
Does closing the Little Mungbean window stop the gateway?
No. The gateway keeps running in the system tray after the window closes so connected clients can continue making requests. Choose Exit from the tray menu to stop it completely. Select Start • Connect.
Where are API keys stored?
API keys are encrypted on this device. The gateway listens only on the local loopback address and does not expose a port to your LAN or the internet.
How can I send feedback?
Bug reports and feedback: heimen115@gmail.com
License
This project is licensed under Apache License 2.0.
