Agent Cascade
0
Routes requests from Model Context Protocol clients to locally hosted large language models like Ollama or LM Studio.
소개
Agent Cascade serves as a vital bridge, connecting Model Context Protocol (MCP) clients, such as Windsurf/Cascade, to local language models hosted via platforms like LM Studio or Ollama. This powerful server exposes a chat-completion tool, enabling developers to route AI requests directly to their own local models, bypassing reliance on external hosted APIs. It offers flexibility through configurable environment variables for the local model's base URL and default model, and even supports advanced patterns like self-reflection where a model can "ask itself" for sub-calls.
주요 기능
- Provides a `local_chat` MCP tool for chat completions
- Supports `local.chat` as a direct request handler for custom MCP methods
- Configurable base URL for local LM Studio/Ollama-compatible APIs
- Allows setting a default model for calls where none is specified
- Enables self-reflection and same-model sub-calls with controlled budgets
- 0 GitHub stars
사용 사례
- Integrating local LLMs (Ollama, LM Studio) with Windsurf/Cascade and other MCP clients
- Implementing self-ask or reflection patterns with local models for multi-step agent workflows
- Developing AI applications that leverage private or custom-tuned local language models