Shieldapi MCP

What alberthild/shieldapi-mcp MCP server does

The alberthild/shieldapi-mcp MCP server provides security-oriented tools for agents handling potentially unsafe content or infrastructure. It can inspect URLs for malware and phishing indicators, evaluate domains and IP addresses using reputation and network signals, and check email addresses or SHA-1 password hashes against breach data associated with Have I Been Pwned. The full_scan tool combines relevant checks for a URL, domain, IP address, or email target.

It also covers AI-specific risks. check_prompt analyzes text for prompt injection across direct instruction overrides, encoding tricks, exfiltration attempts, and indirect injection patterns. scan_skill statically reviews an AI skill or plugin for eight risk categories, including malicious code, credential handling, secrets, suspicious downloads, dependencies, and financial access. check_mcp_trust returns a composite assessment of an MCP endpoint using security, injection, supply-chain, reliability, and on-chain reputation signals.

How it works

The MCP process runs locally and forwards tool requests to the ShieldAPI API. Calls can run in demo mode without a wallet, subject to daily per-endpoint limits. The README describes 10 free calls per day for most endpoints, with lower limits for some operations such as password-prefix checks, full scans, and skill scans.

For paid access, the process uses the x402 payment flow. When the API requests payment, the server can use USDC on Base from the configured EVM wallet and then return the requested security result. Listed paid prices range from $0.001 to $0.02 per request, depending on the endpoint.

Setup and configuration

Install and run the server with:

npx shieldapi-mcp

Claude Desktop and Cursor can configure the same command with -y and the package name shieldapi-mcp. Leaving the wallet variable unset selects the free demo mode. Paid usage requires USDC on Base and the wallet's EVM private key.

Supported configuration variables include:

• SHIELDAPI_URL: API base URL; the documented default is https://shield.vainplex.dev.
• SHIELDAPI_WALLET_PRIVATE_KEY: EVM private key used for x402 payments. It is optional for demo access.

Tools and capabilities

• check_url: checks URL malware and phishing risk using URLhaus and heuristic analysis.
• check_password and check_password_range: perform HIBP breach checks for SHA-1 hashes or hash prefixes.
• check_domain: examines DNS, blacklists, SPF, DMARC, and SSL signals.
• check_ip: checks blacklists, Tor exit-node status, and reverse DNS.
• check_email: checks whether an email address appears in known breaches.
• full_scan: combines applicable checks for one target.
• check_prompt: returns an injection verdict, confidence, matched evidence, and decoded content when obfuscation is found.
• scan_skill: reports a 0–100 risk score and severity-ranked findings with file locations; it does not execute code.
• check_mcp_trust: produces a 0–100 trust score, tier, and per-signal breakdown for an MCP endpoint.

Limitations and notes

Results depend on the external ShieldAPI service and its underlying data sources. The skill scanner performs static analysis only, so it does not observe behavior by executing the submitted code. Free access is rate-limited by endpoint and does not provide unlimited usage. Paid requests require a funded wallet, USDC on Base, and safe handling of the configured private key. The documented prompt detector is intended for inline screening before processing untrusted text and reports a confidence score rather than a guarantee of correctness.

The alberthild/shieldapi-mcp MCP server is released under the MIT license. Its API base URL, x402 discovery document, and OpenAPI document are documented in the project README for clients that need direct service information.

For agents evaluating whether to connect to another MCP service, the alberthild/shieldapi-mcp MCP server's trust check is the relevant tool: it combines endpoint security, tool-description injection analysis, supply-chain signals, reliability, and AgentProof ERC-8004 reputation into one assessment. That score should be treated as an input to a connection decision, not as proof that an endpoint is safe.