Think

Implements Anthropic's "think" tool pattern as an MCP server to give Claude and other LLMs explicit reasoning steps during complex workflows. The core tool takes a thought string and logs it without side effects, useful for processing tool outputs, checking policy compliance, or planning multi-step sequences. Standard mode exposes just the think tool via uvx, while advanced mode adds criticize, plan, and search tools (requires Tavily API key). This is the structured reasoning approach Anthropic documented for improving TAU-bench scores in models without native reasoning capabilities. Minimal implementation built on mcp[cli], designed to slot into any agent configuration that supports MCP.