Mund
Scan agent prompts, tool outputs, and code paths for secrets, PII, prompt injection, and data exfiltration before you ship AI features.
Overview
mund is an MCP server for the Ship phase that scans AI-related content for secrets, PII, prompt injection, and exfiltration risks.
What is this MCP server?
- Detect secrets, PII, prompt injection, and exfiltration patterns in AI pipelines
- Configurable MUND_SEVERITY_THRESHOLD (low, medium, high, critical)
- MUND_LOG_LEVEL for debug through error verbosity
- npm package @weave_protocol/mund v0.1.9 with stdio transport
- Part of Weave Protocol; MCP tool, not a markdown skill
- Server version 0.1.9
- 4 detection themes documented (secrets, PII, prompt injection, exfiltration)
- 2 optional environment variables (MUND_SEVERITY_THRESHOLD, MUND_LOG_LEVEL)
What problem does it solve?
Agent apps silently leak API keys, ingest jailbreaks, or forward sensitive data because nothing screens tool I/O at dev time.
Who is it for?
Solo builders shipping LLM agents or RAG apps who want in-IDE guardrails without building a custom safety layer first.
Skip if: Teams needing full SOC2 pen tests, binary SAST, or compliance sign-off from a certified scanner alone.
What do I get? / Deliverables
You get severity-ranked findings in the agent session so you can block or fix risky prompts and outputs before launch.
- Security findings ranked by configured severity threshold
- Coverage for secrets, PII, injection, and exfiltration patterns
- Debuggable scan logs via MUND_LOG_LEVEL
Recommended MCP Servers
Journey fit
How it compares
AI-focused MCP security scanner, not a generic dependency audit or WAF replacement.
Common Questions / FAQ
Who is mund for?
Developers building AI agents with MCP who need quick checks for secrets, PII, injection, and exfiltration during ship and hardening.
When should I use mund?
Use it when reviewing prompts, retrieved docs, or generated code before release, or when debugging suspicious agent behavior.
How do I add mund to my agent?
Install @weave_protocol/mund from npm, optionally set MUND_SEVERITY_THRESHOLD and MUND_LOG_LEVEL, and register the stdio MCP server in your client config.