README.md

metadata

title: HF Papers Search
emoji: 📚
colorFrom: purple
colorTo: blue
sdk: docker
app_port: 7860
short_description: Fast-agent MCP server for Hugging Face Daily Papers search

Community Tool: Production Agents + Eval Workspace

This repo combines two things in one place:

1. Production-facing agents/tools for:
   • Hugging Face Hub Community API workflows
   • Hugging Face Daily Papers search
2. Evaluation assets:
   • benchmark prompts
   • scoring scripts
   • generated reports/plots
   • interpretation notes

If you're new, start with docs/SPACE.md for a guided map.

For generated metrics and reports, use docs/RESULTS.md.

---

Quick start

1) Run the production community agent

fast-agent go \
  --env .fast-agent \
  --agent-cards .fast-agent/tool-cards \
  --agent hf_hub_community \
  -m "show recent activity for huggingface org"

2) Run the production papers search agent

fast-agent go \
  --env .fast-agent \
  --agent-cards .fast-agent/tool-cards \
  --agent hf_papers_search \
  -m "find diffusion papers this month"

---

Repo layout

• Production cards/tools

  • .fast-agent/tool-cards/hf_hub_community.md
  • .fast-agent/tool-cards/hf_api_tool.py
  • .fast-agent/tool-cards/hf_paper_search.md
  • .fast-agent/tool-cards/hf_papers_tool.py

• Eval prompts + expected labels

  • scripts/hf_hub_community_challenges.txt
  • scripts/hf_hub_community_coverage_prompts.json
  • scripts/tool_routing_challenges.txt
  • scripts/tool_routing_expected.json

• Eval runners/scorers

  • scripts/score_hf_hub_community_challenges.py
  • scripts/score_hf_hub_community_coverage.py
  • scripts/eval_hf_hub_prompt_ab.py
  • scripts/score_tool_routing_confusion.py
  • scripts/run_tool_routing_batch.py
  • scripts/eval_tool_description_ab.py
  • scripts/plot_tool_description_eval.py

• Eval outputs + analysis docs

  • docs/hf_hub_community_challenge_report.{md,json}
  • docs/tool_routing_eval/*
  • docs/tool_description_eval/*
  • docs/tool_description_eval_setup.md
  • docs/tool_description_eval/tool_description_interpretation.md

---

Notes

• Eval scripts resolve repo paths relative to the script location.
• You can run them from any clone path without editing hardcoded roots.

Run all evals

scripts/run_all_evals.sh