Roadmap

Where MCP Skills is headed. We build in public and ship weekly.

Live Shipped and running

14-signal trust scoring

Static analysis across 4 dimensions (Alive, Legit, Solid, Usable) with Skills Mode for MCP servers and AI agent skills.

MCP server + API

Score repos from Claude Code, Cursor, or any MCP client. Agent-optimized API with go/no-go decisions. Free and Pro tiers.

Nightly crawl + daily scan

Automated discovery of new MCP servers from registries and GitHub. Daily re-scoring of monitored repos with email alerts.

Audit logging

Every API call logged with privacy-safe IP hashing, API key prefix, result tier, and timestamp. Daily log rotation.

Certified Safe badge program

Semi-automated certification for repos meeting trust thresholds. Gold badge for READMEs, fast-path trust for agents.

Building In progress

Weekly security digest

Automated analysis of all scored repos: tier trends, score changes, new discoveries, security concerns. Published weekly.

Registry-scale scanning

Score every server in the MCP Registry and major marketplaces. Target: 500+ repos with nightly freshness.

Planned Next quarter

Tool call policy enforcement

Define and enforce policies for which tools an AI skill can call. Block dangerous tool combinations before they execute.

CI/CD integration

GitHub Action and CLI tool to block untrusted skills in pull requests. Enforce trust thresholds before merge.

Comparison pages

Head-to-head trust comparisons (e.g., "Vigile AI vs MCP Skills") for SEO and buyer education.

Exploring Research phase

Runtime behavior monitoring

Sandbox execution to detect context leaks, dynamic payload fetching, and multi-step agent chain attacks that static analysis can't catch.

OpenClaw integration

Parse SKILL.md YAML frontmatter for security transparency scoring. Reward repos that declare permissions and credential scopes.

Have feedback or feature requests? Email hello@mcpskills.io or open an issue on GitHub.