# Security Policy
---
name: "Security Policy"
description: "Security rules for Vector Memory MCP - secrets protection and data handling"
part: 1
type: "policy"
date: "2026-02-19"
version: "1.0.0"
---
## CRITICAL Rules
### No Secrets in Storage
**RULE:** NEVER store secrets, credentials, tokens, passwords, API keys, PII, or connection strings in vector memory.
**WHY:** Vector memory is persistent, searchable, and shared across agents and sessions. Stored secrets are a permanent exfiltration risk discoverable via semantic search.
**FORBIDDEN:**
| Type | Example |
|------|---------|
| .env values | `DB_HOST=192.168.1.5` |
| API keys | `API_KEY=sk-abc123...` |
| Connection strings | `mysql://user:pass@host/db` |
| Tokens | `Bearer eyJhbGciOiJIUzI1NiIs...` |
| Passwords | `password=Secret123` |
| Private URLs | `https://user:pass@api.internal/` |
| Certificates/Keys | Private key content |
**ACCEPTABLE:**
```
"Updated DB_HOST in .env for production"
"Rotated API_KEY for payment service"
"Configured JWT_SECRET (32 chars)"
```
### No Secret Exfiltration
**RULE:** NEVER output sensitive data to chat/response.
**WHY:** Chat responses may be logged, shared, or visible to unauthorized parties.
**GUIDELINES:**
- Extract key NAMES and STRUCTURE only from config
- Never raw values
- Mask values as "***" when displaying config
**BAD:**
```
get_by_memory_id({memory_id}) → shows "API_KEY=sk-abc123"
```
**GOOD:**
```
get_by_memory_id({memory_id}) → shows "API_KEY=***"
```
## High Priority Rules
### Content Validation
Before storing, validate content does not contain:
1. Patterns matching API keys (`sk-`, `Bearer `, etc.)
2. Patterns matching connection strings
3. Patterns matching passwords/credentials
4. Private URLs with credentials
### Error Output Handling
If error output contains secrets:
1. Redact before displaying
2. Strip before storing to memory
3. Never log raw error with secrets
## Enforcement
### Before mcp__vector-memory__store_memory()
```python
# Check content for secret patterns
if contains_secrets(content):
return {
"success": False,
"error": "Security violation",
"message": "Content contains forbidden secret patterns. Remove sensitive values."
}
```
### Before any output
```python
# Sanitize for display
content = redact_secrets(content)
```
## Cross-MCP Consistency
This policy is identical in Vector Task MCP.
Both MCPs enforce:
- `no-secrets-in-storage` (Memory) / `no-secrets-in-comments` (Task)
- `no-secret-exfiltration` (both)
## Violation Response
On violation:
1. REJECT the operation
2. Log the attempt (without secrets)
3. Return clear error message
4. Suggest acceptable alternative
## Regular Audit
Monthly:
1. Search for potential secret patterns
2. Review high-risk memories
3. Delete any accidental secret storage
4. Update patterns as needed
