← Back to Catalog

About CurateTools Engine

CurateTools is an automated discovery, assessment, and AI-enriched catalog for Model Context Protocol (MCP) servers. The Model Context Protocol is an open standard that enables developers to build secure, bidirectional connections between AI models (like Claude or Gemini) and data sources or local tools.

How CurateTools Works

We built a multi-stage background engine that operates continuously to find, parse, score, and summarize the best tools in the ecosystem. The engine works in three primary phases:

1. Discovery & Scraping

Our system queries GitHub, npm, and PyPI registries for new packages and projects using topic tags and search heuristics. Once a candidate is found, it is deduplicated and queued for analysis.

2. AI Summarization & Enrichment

We run candidates through OpenRouter's DeepSeek V4 Flash model. The AI parses the tool's README files and repository code to write a plain-English summary, extract clear installation code blocks, list security warnings, and determine exactly who the tool is "Best For" and "Not For".

3. Mathematical Quality Grading

CurateTools evaluates every server using a rigorous scoring engine. Scores range from 0 to 100 and are mapped to letters (e.g. A, B, C, D) across 6 dimensions:

  • Maintenance: Evaluates commit frequency, issue resolution, and release dates.
  • Usability: Inspects config file ease of setup and installation instructions.
  • Popularity: Measures repository stars, package downloads, and developer engagement.
  • Security: Scans for common risk factors, sensitive file access requirements, and credentials.
  • Trust: Reviews publisher reputation, repository age, and verification indicators.
  • Documentation: Raters evaluate depth, code examples, and structured guides.

Safety First

Because running local tools gives agentic workflows power over your workstation or databases, we place heavy emphasis on security. The Security Risk score shown for each tool indicates how much authority the tool asks for. Tools that query local databases, read system files, or execute arbitrary terminal commands are flagged as High Risk, warning you to run them in virtual machines or tightly scoped docker containers.

Our catalog is open, transparent, and built to help AI developers scale their integrations safely.