Deploy AI Agents to Production. Without the Audit Panic.
Agent Skills Hub Enterprise audits the open-source MCP servers and agent skills your team deploys, before they reach production. It is for platform, security and compliance teams shipping AI agents. Start with a free audit: tell us your stack and we email the top graded risks within 48 hours. Detection runs on 44 published rules, and the catalog is re-graded every 8 hours.
Deploy-time security scanning, sandbox validation, and audit-ready compliance evidence for every MCP server and agent skill you ship — mapped to SOC 2, ISO/IEC 42001, and the EU AI Act.
Get a free security audit → · Browse the full catalog free · How we grade
One layer between your developers and production
- Pre-deployment Sandbox — Run each skill against de-identified, production-shaped data in isolation. Red-team probes. Kill switch on first anomaly — before it ever reaches a customer.
- Compliance Evidence Pack — Auto-generated PDF: SOC 2 control mapping, ISO/IEC 42001 alignment, EU AI Act risk classification, training-data isolation proof. Hand it straight to your auditor.
- Full-chain Audit Logs — Every tool call, every data flow, every error — replayable in one click. JSON export to your SIEM (Splunk, Datadog, Elastic). 90-day retention by default.
- SSO/SCIM + Fine-grained RBAC — Okta, Auth0, Azure AD ready. Skill-level permissions (which agent calls which tool, with which scope). Engineering-manager approval gate before prod.
How it works
- Connect your existing MCP setup — Point Hub at your registry of MCP servers and agent skills — GitHub, internal mirror, or direct upload. Onboarding takes ≤ 30 minutes.
- Hub runs a static scan + sandbox test — Each skill is statically scanned for prompt injection, credential exposure, and sandbox-escape risk, then runtime-tested against red-team probes in isolation.
- Get a compliance evidence package — Auditor-ready PDF: control mapping, risk classification, audit-log samples, training-data isolation proof. Ship it to procurement, pass review, deploy.
License, SBOM & on-prem mirror
- License compliance matrix — Every skill tagged MIT / Apache / GPL / commercial-restricted, with applicability flags for GDPR / HIPAA / regional regimes — so legal can clear it fast.
- SBOM export — Software bill of materials per skill, plus continuous supply-chain monitoring with webhook alerts when a dependency picks up a CVE.
- Private on-prem mirror — An internal mirror of the audited subset, synced from Hub every 8 hours. Your agents pull skills without reaching the public internet.
Pricing
Free to discover & assess. Paid to validate, govern & prove.
- Free — $0 Forever · no card: Browse the full skill catalog · Quality score + per-dimension breakdown · Basic security scan per skill · Scenario & comparison pages · Newsletter (weekly trending)
- Pro — $29/yr · early supporter (standard $49/yr): Everything in Free · Full README-text deep search · 200 results/page · CSV/JSON export · CLI & MCP API key · Email support
- Team CI Gate — $49/mo · 3–20 engineers · self-serve: Everything in Pro (for your team) · GitHub Action PR skill & MCP gate · 43 SlowMist-aligned security checks · Repo allowlist policy (.agentskillshub-allowlist.json) · 8-hour upstream change / risk alerts
- Enterprise — $10K–$30K /year / department: Everything in Team CI Gate · SSO/SCIM (Okta, Auth0, Azure AD) · Pre-deployment sandbox · Compliance evidence pack · Full-chain audit logs + SIEM export · License matrix · SBOM · on-prem mirror · Dedicated account manager + 4h SLA · Regulated-industry add-ons
Enterprise scales with team size and skills audited. Starter: 10 engineers / 50 skills ($10K/yr). Standard: 50 / 200 ($20K/yr). Business: 200 / unlimited ($30K/yr). 200+ engineers or regulated industries start at $50K/yr.
Procurement FAQ
How is this different from Snyk or SonarQube?
Snyk and SonarQube scan your code. Hub scans the agent skills and MCP servers your code calls into — the supply-chain layer they don't cover. We focus on AI-specific risks: prompt injection, sandbox escape, credential leakage in tool definitions, and model training-data isolation.
Do you train models on our data?
No. Hub does not train, fine-tune, or send your data to any model provider. All analysis runs in our infrastructure or on-prem (Enterprise plan). Data isolation is a contractual term.
EU AI Act — do you cover the full checklist?
Yes. Hub's compliance evidence pack maps to Article 9 (risk management), Article 10 (data governance), Article 11 (technical documentation), and Article 12 (record-keeping). Updated for the August 2026 full enforcement.
Is Hub itself SOC 2 audited?
Not yet — we're honest about that. We are scoping SOC 2 Type II (target Q3 2026). For customers who require SOC 2 today, we provide our security control mapping under NDA, can complete a custom security questionnaire, and will commit to a timeline as part of contract negotiation.
Can we deploy on-prem or private cloud?
Yes, for Enterprise. We support AWS, GCP, Azure, and on-prem Kubernetes, plus a private skill mirror synced every 8 hours. Air-gapped deployments available for healthtech / defense customers.
What happens if a skill is updated upstream?
Hub re-scans on every commit to the upstream repo. Updated skills enter the pre-deployment sandbox automatically, and your engineering team is notified before any auto-pull to production.
How long does a POC take?
Standard POC: 4 weeks. Week 1 onboarding + connector setup. Weeks 2–3 audit your current skill inventory and deliver findings. Week 4 review with your platform team. POC fee: $5,000, fully credited toward an annual plan.
Are you a skill creator, not a buyer?
The Verified Creator program certifies serious skill authors — audited, trusted, and surfaced above the noise of the full catalog. Apply for Verified Creator →
- Authenticated ✓ badge on your skill pages and cards
- Featured slot in Verified Organizations (rankings stay objective, never paid)
- Priority indexing + expedited security & quality audit
- “Contact author” channel so users can reach you directly