Little Mungbean • A free, open-source, simple, lightweight AI gateway manager that gives Codex and Claude clients a shared local gateway that is easy to switch and monitor. It supports multiple providers, local protocol conversion, model mapping and injection, MCP Server and Skills management, usage statistics, and more...
Little Mungbean is simple, with no complicated concepts. Even beginners can understand it at a glance. This documentation is mainly for search engines, so you don't need to spend any more time reading.
Install an Agent client • Little Mungbean supports one-click installation of Codex CLI, Codex Desktop, Claude Code, and Claude Desktop. The example below shows Codex Desktop.
Enter an API endpoint and API key • Enter an official provider's or a third-party relay's API endpoint and API key.
Manage providers • Rename, delete, or drag providers to reorder them. You can also choose a provider preset to fill in the API endpoint automatically, and switch between different Agent clients and providers.
Fetch models • Click a model card to enable or disable injection of that model into the Agent client.
If fetching models fails, your network cannot reach the provider's API endpoint. Configure a network proxy on your computer.
Inject a model list • Choose the list of models to make available in the Agent client.
Default model list • Populate the Agent client's built-in model slots for its own model family, excluding models from other families.
Gateway model list • Inject the full gateway model list into the Agent client, including models from other families.
Switch models • Switch models directly in the Agent client. Models from other families may require protocol conversion.
Start • Connect • Little Mungbean automatically starts or restarts the Agent client and injects the model list.
If Start • Connect fails, your network cannot reach the provider's API endpoint. Configure a network proxy on your computer.
The Agent client's status light turns green • The Agent client can now send chat requests through the local gateway.
View usage statistics • Hover over a provider record card to view its request count, token usage, and cache hits. Statistics are tracked separately for each provider and each API key within the same provider.
Local protocol conversion • When enabled, local conversion supports four protocols. Little Mungbean recommends whether to turn it on or off.
Protocol conversion in an AI gateway or proxy automatically translates the API format used by an Agent client into a format supported by an upstream model. It converts requests and responses in both directions, including streaming responses.
Local protocol conversion currently supports these main formats:
Enable local protocol conversion when you need to use models from other families, for example:
Switch account mode • Click the account sign-in mode button to switch modes with one click.
Right-click an Agent client • Open its configuration directory or copy its connection token.
Open configuration directory • Quickly open the Agent client's configuration directory.
Copy connection token • Use the Agent client with other third-party tools.
Request log • View Agent clients, providers, models, status codes, tokens (input/cache/output), elapsed time, time to first token, request endpoints, and protocol conversion details in real time.
Runtime log • View detailed logs of each operation and any error messages.
Local MCP Servers • Enable, disable, edit, or delete local MCP Servers.
Add an MCP Server • Manually add an MCP Server using JSON.
Add an MCP Server to target clients • Select one or more target clients, then save.
MCP Server market • Fetch or search for MCP Servers in the market, then select one or more target clients to install them.
Local Skills • Enable, disable, edit, or delete local Skills.
Add a Skill • Choose a directory containing SKILL.md or a ZIP archive, or create a new Skill by editing SKILL.md in the template.
Skills market • Fetch or search for Skills in the market, then select one or more target clients to install a Skill.
