Best Safety AI Skills & MCP Servers

BLACK_WALL MCP server — a pre-action risk check your AI agent calls before any irreversible action (send email, move money, run SQL, delete data).

Git-like undo, revert, and rollback checkpoints for OpenAI Codex CLI. MCP server + hooks snapshot every edit. Restore in one cmd. Free, MIT licensed.

Search the PubChem chemical database for compounds, properties, safety data, bioactivity, cross-references, and entity summaries via MCP. STDIO or Streamable HTTP.

Agentic HubSpot CLI + MCP server. Covers ~all documented public endpoints (1180 across 55+ command domains). Reachability is tier/scope/auth-model gated — UI-only surfaces and deprecated APIs excluded. Enterprise safety rails, self-hosted.

Solana Token Safety API & MCP Server — rug pull, honeypot, and safety analysis for AI agents

The agent eval standard for MCP. Score every agent output for quality, safety, and cost.

ThumbGate self-improving agent governance: thumbs-up/down turns every mistake into a prevention rule and blocks repeat patterns. 36 pre-action checks, budget enforcement, and self-protection for Claude Code, Cursor, Codex, Gemini CLI, and Amp.

Package Intelligence MCP server for AI agents. Stops hallucinated/malicious package installs across 19 ecosystems (npm, PyPI, Cargo, Go, Maven, NuGet, RubyGems, Composer, Pub, Hex, Swift, CocoaPods, CPAN, Hackage, CRAN, Conda, Homebrew, JSR, Julia). 22 to

Use the MCPSkills pre-install trust layer from Claude Code, Cursor, or any MCP client. Accepts GitHub repos, npm packages, Smithery URLs, and OpenClaw skills. 15 signals (incl. OSV/KEV/EPSS vulnerability intelligence), safety scanning, OpenClaw frontmatte

Lightweight pre-commit safety gate for AI agents. Answers 'is this change safe?' with PASS/WARN/BLOCK verdict in seconds. Zero setup, no database.

Metacognitive AI agent oversight: adaptive CPI interrupts for alignment, reflection and safety

MCP server for ThoughtProof — verify AI reasoning with adversarial multi-model consensus

Query FDA data on drugs, food, devices, and recalls via openFDA. STDIO or Streamable HTTP.

MCP server for US consumer product recalls from the Consumer Product Safety Commission — hazards, remedies, and affected products.

Pre-execution governance for AI agents. Intercepts MCP tool calls before execution with deterministic blocking, human-in-the-loop holds, and behavioral drift detection.

Cognitive browser automation that thinks like your users—and helps AI agents navigate too. Simulate real user cognition with abandonment detection, constitutional safety, chaos engineering, and UX friction discovery. Sites that pass CBrowser's cognitive t

Decode VINs, search recalls, complaints, crash ratings, and investigations via MCP. STDIO or Streamable HTTP.

Pre-execution governance for AI agents. Intercepts MCP tool calls before execution with deterministic blocking, human-in-the-loop holds, and behavioral drift detection.

Safety first. Hardware-verified DeFi for AI agents — designed for when the AI can be compromised.

MCP server for Tuteliq — 50 AI-powered tools for child safety, fraud detection, synthetic content forensics, identity verification, grooming, bullying, sextortion, document analysis, and content moderation. Interactive UI widgets for Claude, Cursor, and M

MCP server for controlling Home Assistant lights and managing scenes. Lights only by design for safety.

MVA (Model-View-Agent) framework for the Model Context Protocol. Structured perception packages with Presenters, cognitive guardrails, self-healing errors, action consolidation, and tRPC-style type safety — so AI agents perceive and act on your data det

Argosvix MCP server — AI agent から traces / cost / alert / annotations / eval-criteria / prompts / safety / eval-runs / llm-budget / budget-gate / policy-gate / approvals / members / saved-views / audit-log / projects / account-health / alert-proposals / a

MVA (Model-View-Agent) framework for the Model Context Protocol. Structured perception packages with Presenters, cognitive guardrails, self-healing errors, action consolidation, and tRPC-style type safety — so AI agents perceive and act on your data deter