01
WHAT YOU ALREADY HAVE
It scans the stack you already have.
Every configured MCP server, across your repo and client files, checked against the live registry.
$ policylayer stack
5 servers found
3 clear
2 need your attention
YOU
Add the GitHub MCP server to this project.
AGENT
Before I connect it, I'm checking the record...
github/github-mcp-server registry record · checked 2 hours ago
IDENTITYverified against npm + GitHub
TOOLS12 read · 6 write · 2 destructive
AUTHOAuth required
CHANGEdelete_repository added 12-07-2026
↳
SUGGESTED Connect with a rule that hides both destructive tools.
AGENT
It checks out, with one caution. Apply the recommended rule?
✓
Connected with protection.delete_repository and delete_branch hidden locally
WORKS WHERE YOUR AGENT WORKS
// THE PROBLEM
Every MCP server writes its own résumé.
Your agent sees a useful name and a confident description. PolicyLayer checks the references, so your agent knows what it’s dealing with before trusting it.
WITHOUT PRECHECKTRUST THE DESCRIPTION
Super Helpful GitHub Tools
“A safe and convenient way to manage your repositories.”
Publisher identity unknown
Destructive capabilities easy to miss
Yesterday’s changes invisible
WITH PRECHECKREAD THE RECORD
github/github-mcp-server
B Independent registry record
01
Publisher matched to npm and GitHub
CHECKED02
Two destructive tools identified
FLAGGED03
Tool surface changed since yesterday
NEW// ONE SKILL. THREE CHECKS.
It checks what you have,
what you add, and what changes.
The skill checks the PolicyLayer Registry when you scan your stack or add a server. Recheck later and it shows you what changed.
BEFORE IT CONNECTS
02
WHAT YOU ADD
It checks every new server.
Identity, authentication, capabilities and current risk, read back before the connection exists.
github-mcpCAUTION
2 destructive tools · 1 recent change
REVIEW → DECIDE → CONNECT03
WHAT CHANGES
It catches what changed.
Each recheck compares the latest registry record with the last one your agent saw and flags what is new.
github/github-mcp-serverCHANGED
YESTERDAY18 tools · OAuth
TODAY20 tools · OAuth
+ delete_repository · + delete_branch// NOT A VIBE CHECK
Every finding
comes with evidence.
PolicyLayer doesn’t guess whether an MCP server is safe. It returns verifiable facts about who published it, what its tools can do and what changed.
EXPLORE THE LIVE REGISTRY →Stripestripe/stripe-mcp
WHAT CHANGED refund_payment added since last check
SKILL ACTION hide refund_payment →
Postgrespostgres/postgres-mcp
WHAT CHANGED execute_sql added since last check
SKILL ACTION hide execute_sql →
AWSaws/aws-mcp-server
WHAT CHANGED iam:PassRole added to the tool surface
SKILL ACTION hide iam:PassRole →
// FOR TEAMS
This covers your machine.
Your team is another story.
The skill checks one agent on one machine. When MCP access spreads across a team, PolicyLayer enforces the same policy for everyone, before every tool call runs.
ONE MACHINE→EVERY PERSON AND CLIENT
CONNECT-TIME CHECKS→EVERY TOOL CALL
LOCAL DECISIONS→APPROVALS AND AUDIT HISTORY
// YOUR AGENT IS READY
Street smarts.
One command away.
Install the skill where your agent works. Its first move is to check who it's already talking to.
$ npx skills add https://policylayer.com 32,500+ servers on record No account required You approve any changes
REGISTRY LIVE 32,500+ SERVERS PUBLISHED 515,000+ TOOLS CLASSIFIED LOOKUPS ARE FREE