Meta AI + MCP + API connector security

Before the agent connects, find out what you actually exposed.

Connector Security Preflight maps the agent-callable surface before platform review. Use it before Meta AI Connector review—or for any MCP/API agent integration—to enumerate tools and endpoints, identify state-changing and privileged actions, inspect auth and OAuth signals, flag tool-description injection, and calculate blast radius without invoking business actions.

Passive contract inspection No credentials accepted Independent
Run the preflight

Inspect a public connector endpoint.

Use a public HTTPS MCP endpoint, a public OpenAPI JSON URL. The public preflight does not accept OAuth tokens, API keys, passwords, cookies, or customer data. Authenticated enforcement belongs in a scoped assessment.

Connector type

Private, loopback, local, link-local, non-HTTPS, credential-bearing, and nonstandard-port targets are rejected.

Safe boundary: live MCP scanning performs protocol discovery/negotiation and tools/list. It does not call business tools. OpenAPI scanning fetches the contract only. This is independent preflight evidence, not Meta certification.

Nothing has been sent.

Built for the new connector layer

Meta AI is one target. The security problem is bigger than Meta.

The same MCP server or API can become agent-callable across multiple ecosystems. The preflight evaluates the exposed contract itself, so the evidence remains useful as connectors move between agent platforms.

Left Out Security is independent and is not affiliated with, endorsed by, or certifying on behalf of Meta or any other connector platform.