What could these AI tools do?
This reviews only what a tool manifest describes. It cannot inspect effective permissions, credentials, approvals or what really happens when a tool runs. It never calls an AI agent or tool.
Check an MCP tool list
Paste a JSON array of tools, a {"tools":[...]} object, or an MCP tools/list response. Your manifest is processed in your browser, not uploaded by this check.
Use tool names, descriptions and schemas only. Do not paste API keys, passwords, private customer records or secrets. If you accept analytics cookies, we count that a check finished and roughly how many tools it had, never the manifest itself.
What the manifest describes
Not established for any tool: what the agent or its credentials can really do, which approvals or controls apply, and what happens when a tool runs. These are possibilities to discuss, not security findings.
Was this useful?
What was missing?
What you can do next
Each of these is free and runs on your machine. None of them needs an AtlaSent account.
Make risky tool calls wait for you
In Claude Code: /plugin marketplace add AtlaSent/atlasent-mcp-server, then /plugin install atlasent-guard@atlasent.
Other MCP hosts
MCP Gate puts the same fail-closed check in front of the tools of Cursor, Windsurf and other hosts.
Check your deploy pipeline
List the deploy, publish and migration steps in your CI that nothing can stop: npx -p @atlasent/mcp-server atlasent-exposure-scan .
Enforcing an action in a real organization needs separate setup and verification. How AtlaSent works with AI agents.
Other free checks: Change Check · Approval Tester · Change control self-check · all free tools