Security

Security is always our #1 priority. If a feature cannot be implemented safely, it will not be implemented.

---

10 Non-Negotiable Guarantees

Every qso-graph MCP server guarantees:

#	Guarantee	How It's Verified
1	Credentials never in logs	Static analysis + code review
2	Credentials never in tool results	All return paths traced
3	Credentials never in error messages	All raise/exception statements audited
4	Credentials never in config files	OS keyring only (via adif-mcp)
5	No command injection surface	No subprocess, no shell=True, no os.system
6	No SQL injection surface	Parameterized queries or API-only
7	HTTPS only for external calls	All URLs audited
8	Rate limiting implemented	Prevents user account bans
9	Input validation on all user strings	Regex validation before use
10	Fail safely	Errors reveal no sensitive information

---

Credential Architecture

Credentials flow through the OS keyring and never appear in the MCP protocol:

User runs: adif-mcp creds set --persona ki7mt --provider qrz --password ****
                ↓
        OS Keyring (encrypted by OS)
        - macOS: Keychain
        - Windows: Credential Manager
        - Linux: Secret Service (GNOME Keyring / KWallet)
                ↓
User asks: "Look up W1AW on QRZ"
                ↓
AI calls: qrz_lookup(persona="ki7mt", callsign="W1AW")
                ↓
qrz-mcp: reads credentials from keyring (in-process only)
                ↓
qrz-mcp: HTTPS request to api.qrz.com
                ↓
AI sees: {"call": "W1AW", "name": "ARRL", "grid": "FN31pr", ...}
                ↓
Password NEVER appears in: tool params, tool results, logs, errors

What the AI Can See

• Tool parameters: persona, callsign, band, mode, etc.
• Tool results: lookup data, QSO records, statistics
• Error messages: static strings like "Login failed" (never "Login failed with password X")

What the AI Cannot See

• Passwords
• API keys
• Session tokens
• Any credential values

This is enforced by architecture, not policy. Credential values physically do not exist in MCP protocol messages.

---

Input Validation

All user-provided strings are validated before use:

import re

CALLSIGN_RE = re.compile(r'^[A-Z0-9]{1,3}[0-9][A-Z0-9]{0,4}(/[A-Z0-9]+)?$', re.IGNORECASE)
GRID_RE = re.compile(r'^[A-R]{2}[0-9]{2}([A-X]{2})?$', re.IGNORECASE)
DATE_RE = re.compile(r'^\d{4}-\d{2}-\d{2}$')
BAND_RE = re.compile(r'^\d{1,3}[mM]$')
MODE_RE = re.compile(r'^[A-Z0-9]{2,10}$', re.IGNORECASE)

Error messages never include the invalid input value — this prevents injection via error reflection.

---

Forbidden Patterns

The following are absolutely forbidden in any qso-graph server:

# FORBIDDEN — Command injection
import subprocess
os.system(command)
subprocess.run(command, shell=True)
eval(user_input)
exec(user_input)

# FORBIDDEN — Credential exposure
print(f"Password: {password}")
logging.debug(f"API key: {api_key}")
raise ValueError(f"Auth failed with password {password}")

# FORBIDDEN — Plaintext credential storage
with open("config.json") as f:
    password = json.load(f)["password"]

---

Release Process

No package is published to PyPI without completing the full security pipeline:

1. Development — implement feature, run local security checklist
2. CI Pipeline — pip-audit, security test suite, static analysis
3. Security Audit Request — developer sends request via internal message queue
4. Security Review — independent agent reviews all source files, credential flows, error messages, input handling, and dependencies
5. Release Decision — PASS required before tagging and publishing

---

Security Test Suite

Every server includes automated security tests:

• No print statements containing credential keywords
• No logging statements containing credential keywords
• No subprocess or shell execution
• All external URLs use HTTPS
• No credentials in exception messages

---

Incident Response

Severity	Description	Response Time
Critical	Credential exposure, RCE	Immediate (hours)
High	Data leakage, auth bypass	24 hours
Medium	Information disclosure	1 week
Low	Minor issues	Next release

Reporting Vulnerabilities

Do NOT open public GitHub issues for security vulnerabilities.

Email: ki7mt@yahoo.com Subject: [SECURITY] qso-graph vulnerability report

Include: - Affected package and version - Steps to reproduce - Potential impact