Open-source platform scores 39,752 MCP servers across three dimensions

Agent Tool Intelligence built and open-sourced a scoring platform that evaluated 39,752 MCP servers, publishing its findings on Dev.to. The team's first attempt used static analysis alone — covering schema correctness, token efficiency, description quality, security, and install reliability — but the result was a near-useless distribution where 85.7% of all tools received a Grade B. With no differentiation, the platform provided no incentive for improvement, prompting a full rebuild.

A "Quality Floor" cap prevents popularity from outranking engineering quality — a tool with 10,000 stars but poor construction cannot exceed its floor.

The new composite scoring model combines three independent dimensions: a Quality Score (0–100) based on static analysis weighted across Schema (25%), Token Efficiency (25%), Description (20%), Security (15%), and Install (15%); a Community Bonus (0–60) derived from GitHub stars on a log scale, activity recency, and official/verified status; and a Trust Bonus (0–30) based on real execution data, where no data yields no bonus rather than a penalty. A "Quality Floor" cap prevents popularity from outranking engineering quality — a tool with 10,000 stars but poor construction cannot exceed its floor.

The resulting grade distribution is more revealing: 54.0% of tools sit at C (good foundation, no community signal), 18.1% at B, 13.8% at C+, 10.9% at D, 2.8% at B+, and 0.4% at F. The platform surfaces a clear path for improvement — reaching B from C requires 10 stars, while reaching B+ requires 50 stars plus recent activity. Developers can score their own tools by submitting a GitHub repo URL and embed a live grade badge in their README. The project is MIT-licensed.

Key facts

1. 0139,752 MCP servers were scored by the Agent Tool Intelligence platform.
2. 02The original single-dimension scoring model gave 85.7% of all tools a Grade B, prompting a full rebuild.
3. 03The new composite grade combines a Quality Score (0–100), Community Bonus (0–60), and Trust Bonus (0–30).
4. 04Quality Score weights: Schema 25%, Token Efficiency 25%, Description 20%, Security 15%, Install 15%.
5. 0554% of tools score Grade C — solid quality but zero community adoption, making them invisible to agents.
6. 06A Quality Floor cap prevents a high-star tool with poor engineering from exceeding its quality-based ceiling.
7. 07The platform is open source under the MIT license and provides embeddable grade badges for GitHub repos.

Topics