Risk Scoring

Risk Scoring

MCPShield calculates a risk score (0–100) for each discovered MCP server based on its configuration.

Risk Levels

Level	Score Range	Alert Generated?
Critical	85–100	Yes — immediate
High	60–84	Yes — immediate
Medium	30–59	No (score increase > 10 triggers one)
Low	0–29	No

---

Scoring Factors

Signal	Weight	Examples
Shell / command execution	+35	bash, sh, exec, run_command
High-risk tool names	+25	delete_file, drop_table, execute_query
Docker / container access	+20	docker run, container tool names
Sensitive path access	+20	/, C:\, /etc, /root
Database access	+15	postgres, mysql, mongodb connection strings
Filesystem write tools	+15	write_file, create_file, delete
Sensitive env var names	+15	DATABASE_URL, POSTGRES_PASSWORD, API_KEY
Network / HTTP tools	+10	fetch, http_request, outbound calls

Scores are capped at 100. Multiple signals stack.

---

Example Calculations

postgres-prod (Critical)

Shell access detected        +35
Sensitive env var (DB_URL)   +15
Database access pattern      +15
High-risk tool (execute_sql) +25
─────────────────────────────────
Total: 90  →  Critical

filesystem-home (High)

Filesystem write tools       +15
Sensitive path (/home)       +20
Network tools                +10
Sensitive env (HOME, PATH)   +15
─────────────────────────────────
Total: 60  →  High

github-api (Low)

Network / HTTP tools         +10
Sensitive env (GITHUB_TOKEN) +15
─────────────────────────────────
Total: 25  →  Low

---

Alert Rules

Alerts are generated automatically when:

• A new server is discovered with risk score ≥ 60
• An existing server's risk score increases by more than 10 points between scans

---

Customization

Scoring weights are defined in:

backend/app/utils/risk_scorer.py

Modify the weight constants there to tune sensitivity for your environment.