mcp tool safety schema gatekeeper
by nowrich
Production-grade safety and schema auditor for Model Context Protocol (MCP) tool definitions.
Secure checkout via Stripe
Works with the AI tools you already use
See it in action
You say
Review this MCP tool: delete_record(id: string, force: boolean). It deletes a row from the production DB without confirmation if force is true. Response is just a string.
Your agent does
STATUS: BLOCKED Findings:
- Destructive Actions: FAIL. Must require a typed confirmation_text like "delete-production-db" instead of a simple boolean.
- Response Shapes: FAIL. Returns plain string; must use a standard JSON envelope.
- Validation: FAIL. Missing minLength for id.
What you get
About this skill
The problem
Deploying Model Context Protocol (MCP) tools with loose schemas or inadequate permission checks leads to prompt injection risks and catastrophic data loss. Developers often forget to implement explicit confirmations for destructive actions or fail to validate response shapes, causing integration breaks.
What it does
- Enforces strict JSON Schema validation for all tool parameters to prevent malformed payloads and denial-of-service attacks.
- Audits permission scopes to ensure tools follow the principle of least privilege and do not leak environment secrets.
- Mandates explicit user confirmation flows for destructive operations like database drops or file deletions.
- Verifies response shape consistency to ensure predictable output for downstream LLM consumption.
- Checks for naming collisions and ambiguous identifiers that confuse agentic reasoning.
Frameworks & tools
Model Context Protocol (MCP), JSON Schema, Node.js, Python, and TypeScript MCP SDKs.
Why this beats prompting it yourself
Writing a prompt to check safety often misses edge cases like additionalProperties: false or specific enum constraints. This skill provides a systematic, multi-point audit trail that mimics a professional security review, ensuring nothing is overlooked before production deployment.
Use cases
- Reviewing a new MCP server manifest before registering it with Claude or Cursor.
- Auditing destructive tools like 'delete_row' to ensure they require a manual confirmation flag.
- Refactoring legacy tool schemas to meet strict JSON Schema standards.
- Generating a safety compliance report for team lead approval before a production release.
Known limitations
Does not perform runtime traffic analysis or live penetration testing. Requires the developer to provide the tool definitions or source code for analysis.
How to install
Drop the file into your AI Agent. Works with Claude, Cursor, ChatGPT, and 20+ more.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill passes our trust checks below.
Security scanned
Passed our 8-point scan before listing
Fresh listing
Recently published to Agensi
30-day refund
Not a fit? Get your money back
Trust & safety
Security scanned
Verified clean today
- 30-day refund guarantee
- One-time purchase, yours forever
- Secure checkout via Stripe
Frequently Asked Questions
Popular in Documentation
incident-postmortem
Transform raw incident logs and Slack threads into blameless, structured postmortems and 5-Whys RCA reports.
Developer Documentation Writer
Produces developer documentation structured around what developers are trying to do — information architecture, quickstart, how-to guides, reference docs, gap audit — for the product described. Use when the user says write developer documentation, create an API quickstart guide, write a README for this library, or audit our docs for gaps. Do not use when the user wants end-user help articles or marketing copy.
Agent-safe database access kit: harden Postgres before you let an AI tool near it
Builders wiring LLM agents to their DB keep discovering the hard way that a chatty tool call can drop rows, exfiltrate P
Research-to-Report
Decision-grade web research: tiered sources, real citations, honest limitations. No API keys.