Skip to content

Security: agentra-commerce/xap-sdk

Security

SECURITY.md

Security Policy — XAP SDK MCP Server

Authentication

  • Sandbox mode (XAP_MODE=sandbox): No credentials required. Uses local in-memory settlement engine. No external API calls.
  • Live mode (XAP_MODE=live): Requires XAP_API_KEY environment variable. No default value — the server will not authenticate without an explicit key.

Permissions Explained

env_vars (read-only)

The server reads two environment variables:

  • XAP_MODE — determines sandbox vs live operation
  • XAP_API_KEY — authentication for live mode API calls
  • XAP_API_URL — optional override for API endpoint

No environment variables are written or modified.

process_spawn

Used only in the optional setup.py script to invoke claude mcp add for Claude Desktop configuration. The MCP server itself (server.py) does not spawn any processes during normal tool operation.

HTTP Network Access

In live mode, the server makes HTTPS requests to a single endpoint:

  • https://api.zexrail.com/api/v1/*

No other external endpoints are contacted. In sandbox mode, no network requests are made.

File System Read

The server reads local schema files for validation. No file system writes occur during tool operation.

Data Storage

Tool state (negotiation contracts, verity receipts) is stored in bounded in-memory caches:

  • Maximum 1,000 entries per cache
  • Automatic expiry after 1 hour (TTL)
  • Entries are cleaned up on each tool call
  • No data persists to disk
  • No data is shared between sessions

Dependency Security

All dependencies are pinned to versions that address known CVEs:

  • mcp >= 1.9 — addresses DNS rebinding (CVE-2025-66416), validation DoS (CVE-2025-53366), HTTP transport DoS (CVE-2025-53365)
  • pydantic >= 2.11 — addresses regex DoS (CVE-2024-3772)
  • jsonschema >= 4.23 — latest stable

Reporting Vulnerabilities

Report security vulnerabilities to: security@zexrail.com

Please include:

  • Description of the vulnerability
  • Steps to reproduce
  • Expected vs actual behavior
  • Impact assessment

We aim to respond within 48 hours and patch critical vulnerabilities within 7 days.

There aren’t any published security advisories