MCP Server Security Checklist 2026

Published March 31, 2026 · Updated regularly · By Probe

MCP (Model Context Protocol) servers are becoming the standard way AI agents connect to tools, data sources, and APIs. But most MCP servers ship without basic security and compliance checks — leaving them vulnerable to attacks, regulatory fines, and broken agent integrations.

This checklist covers the 34 checks every MCP server should pass before going to production. Each check can be automatically tested by Probe in under 30 seconds.

1. Transport & Encryption

• 01 SSL/TLS — Valid HTTPS certificate with 30+ days until expiry Security
• 02 DNSSEC — Domain has DNSSEC enabled to prevent DNS spoofing Security
• 03 CAA Records — Certificate Authority Authorization restricts who can issue certs Security
• 04 DMARC + SPF — Email authentication prevents phishing from your domain Security

2. Security Headers

• 05 Security Headers — HSTS, CSP, X-Content-Type-Options, X-Frame-Options, Referrer-Policy Security
• 06 CORS — Access-Control-Allow-Origin properly configured Security
• 07 Content-Type — JSON endpoints return application/json Security
• 08 Rate Limiting — X-RateLimit-Limit or Retry-After headers present Security

3. API Quality

• 09 Response Time — Average response under 500ms Security
• 10 Error Handling — 404 returns structured JSON, not HTML stack traces Security
• 11 API Versioning — Versioned paths (/v1/) or version headers Discovery
• 12 OpenAPI Spec — /openapi.json with valid 3.x schema Discovery
• 13 Documentation — /docs endpoint accessible Discovery

4. MCP & Agent Discovery

• 14 MCP Endpoint — /mcp or /.well-known/mcp.json discoverable Discovery
• 15 Agent Card — /.well-known/agent.json with name, skills, auth Discovery
• 16 A2A Protocol — Google Agent-to-Agent discovery via agent.json Discovery
• 17 llms.txt — Machine-readable documentation for LLM crawlers Discovery
• 18 robots.txt — AI crawler rules (GPTBot, ClaudeBot, etc.) Discovery
• 19 AI Plugin Manifest — /.well-known/ai-plugin.json Discovery

5. Compliance & Regulatory

• 20 Privacy Policy — GDPR-compliant privacy endpoint Compliance
• 21 Terms of Service — Accessible ToS endpoint Compliance
• 22 AI Disclosure — Model card with risk level, capabilities, limitations Compliance
• 23 Human Oversight — Kill switch / emergency stop (EU AI Act Art. 14) Compliance
• 24 security.txt — /.well-known/security.txt with contact info Compliance
• 25 Status Page — Health endpoint or status subdomain Compliance

6. Payment & Trust (Advanced)

• 26 x402 Protocol — Payment discovery via /.well-known/x402.json Trust
• 27 x402 Compliance — HTTP 402 + payment headers on paid endpoints Trust
• 28 Travel Rule — FATF VASP disclosure for payment services Trust
• 29 Wallet Trust — Wallet address discoverable and not sanctioned Trust
• 30 ERC-8004 — Onchain agent identity registration verified Trust
• 31 Auth Maturity — Authentication scheme quality (OAuth, API keys, x402) Trust
• 32 Endpoints — Live, reachable API endpoints detected Discovery
• 33 OASF Classification — Open Agent Service Framework role declaration Discovery
• 34 MCP Transport Security — HSTS + CORS configured on MCP endpoint Security

How to use this checklist

Option 1: Automated scan (recommended)

Paste your MCP server URL into Probe and get all 34 checks in one report. Each failing check includes a fix guide.

Option 2: CI/CD integration

Add Probe to your deployment pipeline:

curl -s -X POST https://getprobe.xyz/api/audit \
  -H "Content-Type: application/json" \
  -d '{"url":"https://your-mcp-server.com"}' | jq '.score, .grade'

Option 3: One-click fix

For $29 (one-time), Probe's Autofix creates a GitHub PR with all missing files — agent.json, security.txt, privacy policy, and more. It also fixes DNS records (DNSSEC, DMARC, CAA) via Cloudflare API.

Why this matters in 2026

The EU AI Act enters full enforcement on August 2, 2026. AI systems — including MCP servers that process data or make decisions — must demonstrate transparency, human oversight, and risk management. Non-compliance can result in fines up to €35 million or 7% of global revenue.

Beyond regulation, MCP servers that pass these checks are more discoverable by AI agents, more trusted by users, and less likely to suffer security incidents.