If you discover a security vulnerability in cve-mcp, please report it responsibly.

Do not open a public GitHub issue for security vulnerabilities.

Instead, send an email to:

contact@orhanyildirim.us

Please include the following in your report:

• Description of the vulnerability
• Steps to reproduce
• Potential impact
• Suggested fix (if any)

• Acknowledgment: Within 48 hours of receiving the report
• Initial assessment: Within 7 days
• Fix or mitigation: As soon as practically possible
• Public disclosure: 90 days after the initial report, or when a fix is released, whichever comes first

You will be credited in the release notes unless you prefer to remain anonymous.

cve-mcp is a read-only intelligence tool that queries public vulnerability databases. Here is what you should know about its security posture:

• Queries public APIs (NVD, EPSS, CISA KEV, GitHub Advisory, OSV) over HTTPS
• Parses and aggregates vulnerability data
• Communicates with AI agents over stdio (standard input/output) transport only

• No authentication data is stored. API keys (NVD_API_KEY, GITHUB_TOKEN) are read from environment variables at runtime and never written to disk or logged.
• No network listeners. The server uses stdio transport exclusively. It does not open any ports or accept inbound connections.
• No data persistence. All caches (NVD results, KEV catalog) are in-memory only and discarded when the process exits.
• No code execution. The server does not execute any code from vulnerability data or external sources.
• No write operations. The server only reads from upstream APIs. It never modifies, creates, or deletes any external resources.

• API keys are passed exclusively via environment variables (NVD_API_KEY, GITHUB_TOKEN)
• Keys are used only in outbound HTTPS request headers
• Keys are never included in tool responses, logs, or error messages
• Both keys are optional -- the server operates with reduced rate limits without them

The following are considered valid security issues:

• Injection vulnerabilities -- if crafted tool input could lead to unintended behavior
• Data leakage -- if API keys, tokens, or sensitive data are exposed in tool responses or error messages
• Authentication bypass -- if transport-level security assumptions can be circumvented
• Dependency vulnerabilities -- critical vulnerabilities in @modelcontextprotocol/sdk or zod that affect cve-mcp
• Path traversal or file access -- any scenario where the server reads or writes files it should not

The following are not considered cve-mcp security issues:

• Rate limiting from upstream APIs -- NVD, EPSS, GitHub, and OSV enforce their own rate limits. This is expected behavior, not a vulnerability.
• Upstream API availability -- If NVD or another source is down, the server returns an error. This is by design.
• Data accuracy -- Vulnerability data comes directly from upstream sources. Inaccuracies are upstream issues.
• Denial of service via large queries -- Sending extremely broad search queries may result in slow responses. This is a limitation, not a vulnerability.
• Social engineering or phishing -- Attacks that require tricking users into running malicious configurations

• Security reports: contact@orhanyildirim.us
• General issues: GitHub Issues
• Author: Orhan Yildirim