chain-of-draft
以最少token实现高效LLM推理. chain-of-draft is a Model Context Protocol (MCP) server maintained by stat-guy. It connects to MCP-compatible clients such as Claude Desktop, Cursor, Cline, and other agents that speak the protocol. It exposes 7 tools, including chain_of_draft_solve, math_solve, code_solve, logic_solve, get_performance_stats, get_token_reduction and analyze_problem_complexity, that an agent can call directly. It is categorized under AI/Reasoning, Tool/Server and AI/Claude.
什么是MCP
草案链(CoD)是一种LLM推理方法，通过生成简洁的中间步骤（通常≤5词）在保持准确性的同时高效解决问题。与思维链(CoT)相比，它能显著减少token消耗。
如何使用MCP
通过Python或JavaScript实现安装，使用Anthropic API密钥配置，可集成Claude桌面版或直接通过提供的客户端库解决数学、编程、逻辑等各领域问题。
MCP适用场景
CoD特别适合需要推理且注重成本效益的LLM应用场景，包括数学解题、编程辅助、逻辑推理任务，以及任何需要在保持准确性的同时减少token消耗的场合。
AIMCP authority
DR and traffic signal for the AIMCP public domain.
Frequently asked questions
What is the chain-of-draft MCP server?
chain-of-draft is a Model Context Protocol server from stat-guy. It lets MCP-compatible AI clients call its tools over a standard interface, so agents like Claude, Cursor, and Cline can use it without custom integration.
How do I connect chain-of-draft to my AI client?
Add chain-of-draft to your client's MCP configuration using the stdio or SSE connection shown in the usage examples on this page, then restart the client to load the server.
What tools does chain-of-draft provide?
chain-of-draft provides 7 tools: chain_of_draft_solve, math_solve, code_solve, logic_solve, get_performance_stats, get_token_reduction and analyze_problem_complexity.
Is chain-of-draft free to use?
chain-of-draft is listed on AIMCP for free. Any API keys or accounts required by the underlying service are set by its provider.
Vernclaw Plugins for OpenClaw
Ready-to-use connectors for SEO data, social reading & content generation. Pay-as-you-go credits with audit logs.