Scampi is an MCP server designed to integrate with Claude Code, enabling it to leverage external Large Language Models (LLMs) for high-volume, token-intensive tasks. This allows Claude Code to offload bulk processing to cheaper local LLM servers (like LM Studio, Ollama, llama.cpp, text-generation-webui) or cloud providers (Z.ai), reserving Claude's context for higher-level judgment and reasoning. Scampi provides a suite of tools for tasks such as indexing codebases for semantic search, generating multiple solution approaches in parallel, and brainstorming diverse ideas, all while managing concurrency and caching for optimal performance.
Características Principales
01Multi-provider LLM support for local and cloud models
02Auto-detection of running local LLM servers
03Codebase mapping with semantic summaries and incremental caching
04Parallel generation of multiple coding solutions or brainstormed approaches
05Token arbitrage to optimize LLM cost and performance
061 GitHub stars
Casos de Uso
01Generating diverse coding solutions in parallel for complex development tasks
02Indexing and searching large codebases with natural language queries
03Quickly brainstorming multiple architectural approaches or implementation ideas
04Retrieving specific file contents and dependency graph information from a project
