Mcp Zero

What is Mcp Zero

MCP-Zero is a proactive toolchain construction framework designed for Large Language Model (LLM) agents, enabling them to build effective toolchains from scratch for various tasks.

Use cases

Use cases for MCP-Zero include developing intelligent assistants, automating content creation, and enhancing decision-making processes in various industries.

How to use

To use MCP-Zero, users can access the dataset available at ‘MCP-tools/mcp_tools_with_embedding.json’ and utilize the tools defined within to create customized workflows for their LLM agents.

Key features

Key features of MCP-Zero include a comprehensive dataset of 308 servers and 2,797 tools, proactive construction of toolchains, and the ability to embed descriptions for enhanced functionality.

Where to use

MCP-Zero can be used in fields such as artificial intelligence, natural language processing, and any domain requiring automated toolchain construction for LLM applications.

MCP-Zero: Proactive Toolchain Construction for LLM Agents from Scratch

Paper: https://arxiv.org/abs/2506.01056

Aha, you find the repo of MCP-Zero so early! :D

We’ve released our dataset and partial code, while more features are on the way. Please check it out!

Using MCP-Zero to proactively construct toolchains for "Making a great meal"

Method: MCP-Zero

• File Path: ./MCP-zero/

We have now released our code for hierarchical semantic matching, and other features will be added recently. Leave a star🌟 to let me know you are staying updated :D

Dataset: MCP-tools

• Dataset Path: ./MCP-tools/mcp_tools_with_embedding.json

• Google Drive: Download Link

• Huggingface Link: Coming soon

My Git LFS bandwidth quota has been exhausted, please use the Google Drive link to download instead. Thank you all for your attention to this work.

Introduction: A dataset containing all filtered tools from the MCP official repo. 308 servers and 2,797 tools in total.

Data structure:

{
  "server_name": string, // The name of the MCP server, extracted or inferred from the README
  "server_summary": string, // A summary of the server's purpose and capabilities, based on all relevant parts of the README.
  "server_description": string, // Description from metadata. 
  "description_embedding": float[3072], // The embedding of the server description from text-embedding-3-large
  "summary_embedding": float[3072], // The embedding of the server summary from text-embedding-3-large
  "tools": [
    {
      "name": string, // The function/tool name
      "description": string, // A concise description of what the tool does
      "description_embedding": float[3072], // The embedding of the tool description from text-embedding-3-large
      "parameter": { // A dictionary of input parameters, being included if explicitly defined
        "param1": "(type) description1",
        "param2": "(Optional, type) description2"
      }
    }
  ]
}

Citation

Citation makes me happy.

–Shakespeare
(just for fun :D)

@article{fei2025mcp,
  title={MCP-Zero: Proactive Toolchain Construction for LLM Agents from Scratch},
  author={Fei, Xiang and Zheng, Xiawu and Feng, Hao},
  journal={arXiv preprint arXiv:2506.01056},
  year={2025}
}