Skip to content
Back to search

MCP-Zero

MCP-Zero: Active Tool Discovery for Autonomous LLM Agents

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/xfey--mcp-zero/badge.svg)](/listings/xfey--mcp-zero)

HTML

<a href="/listings/xfey--mcp-zero"><img src="/api/listings/xfey--mcp-zero/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md · README · 104 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. ## MCP-Zero: Active Tool Discovery for Autonomous LLM Agents
  2. <div style="display: flex; align-items: center; gap: 10px; margin-bottom: 10px;">
  3. <!-- <img src="assets/robot.png" alt="MCP-Zero Robot" width="24" height="24"> -->
  4. <a href="https://arxiv.org/abs/2506.01056">
  5. <img src="https://img.shields.io/badge/Paper-arXiv-red">
  6. </a>
  7. <a href="https://arxiv.org/abs/2506.01056">
  8. https://arxiv.org/abs/2506.01056
  9. </a>
  10. </div>
  11. Thanks for your attention for MCP-Zero! 🤗
  12. We have now open-sourced the code involved in the paper. We will continue to update our work, explore its application in the industry, and continue to expand this project.
  13. <div align="center">
  14. <img src="assets/fig1.png" alt="MCP-Zero workflow">
  15. <p> Using MCP-Zero to proactively construct toolchains for "Making a great meal"</p>
  16. </div>
  17. ### Method: MCP-Zero
  18. ```
  19. MCP-zero/
  20. ├── experiment_apibank.py # experiments: APIBank
  21. ├── experiment_mcptools.py # experiments: mcp_tools (needle test)
  22. ├── matcher.py # code for similarity matching
  23. ├── prompt_guide/ # prompts for our method
  24. ├── reformatter.py # json formatter for tool description
  25. ├── sampler.py # sampler for selecting target tool
  26. ├── test_cases.jsonl # testcase for the matcher
  27. ├── test_matcher.py # unit test for the matcher
  28. └── utils.py # utils: grid_search
  29. ```
  30. We have now released our code for the paper. The code in the paper implements retrieval capabilities and achieves concrete results in experiments.
  31. In our future work, we are committed to applying MCP-zero to the industry, so other modules still need to be involved, such as the dynamic deployment of MCP servers, the environment deployment for GAIA test, etc. We will continue to improve our work, and thank you all for your attention to this work. Leave a star🌟 to let me know you are staying updated :D
  32. ### Dataset: MCP-tools
  33. - **Google Drive**: [Download Link](https://drive.google.com/file/d/1RjBGU-AGdHdhUABoeYSztbfQlD0hjUBn/view?usp=sharing)
  34. - **Huggingface Link**: Coming soon
  35. - **Put the file at**: `./MCP-tools/mcp_tools_with_embedding.json`
  36. **Introduction**: A dataset containing all filtered tools (308 servers and 2,797 tools in total) from the MCP official repo.
  37. **Data structure**:
  38. ```
  39. {
  40. "server_name": string, // The name of the MCP server, extracted or inferred from the README
  41. "server_summary": string, // A summary of the server's purpose and capabilities, based on all relevant parts of the README.
  42. "server_description": string, // Description from metadata.
  43. "description_embedding": float[3072], // The embedding of the server description from text-embedding-3-large
  44. "summary_embedding": float[3072], // The embedding of the server summary from text-embedding-3-large
  45. "tools": [
  46. {
  47. "name": string, // The function/tool name
  48. "description": string, // A concise description of what the tool does
  49. "description_embedding": float[3072], // The embedding of the tool description from text-embedding-3-large
  50. "parameter": { // A dictionary of input parameters, being included if explicitly defined
  51. "param1": "(type) description1",
  52. "param2": "(Optional, type) description2"
  53. }
  54. }
  55. ]
  56. }
  57. ```
  58. **Build dataset on your own**: If you want to build custom dataset for MCP servers, you may follow the code under the `MCP-tools/build_data` folder.
  59. ```
  60. MCP-tools/
  61. ├── build_data
  62. │ ├── get_server_summary.py # code to extract structural data for MCP server's ReadMe file
  63. │ ├── run_vllm.sh # deploy the Qwen2.5-72B-Instruct model with VLLM
  64. │ └── server_summary.prompt # the prompt for extracting dataset
  65. └── download_data.md
  66. ```
  67. ### Citation
  68. > Citation makes me happy.
  69. >
  70. > --Shakespeare
  71. > ~~(just for fun :D)~~
  72. ```bibtex
  73. @article{fei2025mcp,
  74. title={MCP-Zero: Active Tool Discovery for Autonomous LLM Agents},
  75. author={Fei, Xiang and Zheng, Xiawu and Feng, Hao},
  76. journal={arXiv preprint arXiv:2506.01056},
  77. year={2025}
  78. }
  79. ```

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.