Skip to main content
Feedback

MCP evaluation sandbox

Before you deploy new or third-party MCP servers into production, you can use the sandbox environment in MCP Gateway to test, validate, and inspect server behavior in isolation. The sandbox gives you a safe space. Verify that new MCP servers are secure, stable, and properly configured before making them available to your teams.

The sandbox environment lets you run and evaluate MCP servers without exposing any production systems or sensitive data. You can connect, observe, and test how a server behaves under controlled conditions, then decide whether to approve it for production use.

Once a server passes your review, you can promote it directly into your organization’s custom MCP server registry. Approved teams can start using it safely.

Key capabilities

  • Isolated testing: Test MCP servers in a secure environment that’s fully separated from your production systems.
  • Safe Evaluation: Validate server responses, schemas, and data handling before deployment.
  • Promotion Workflow: Approve validated servers and add them directly to your custom MCP server registry.
  • Secure Key Testing: Use your existing secret management setup to safely test credentials and integrations.

Observability and policy testing with the LLM Gateway

All sandbox traffic can be routed through the LLM Gateway, giving you deep visibility into each server and endpoint before it reaches production. For example, you can:

  • View all sandboxed servers and exposed endpoints.
  • Inspect request behavior, performance, and potential data leaks.
  • Apply and test access or rate-limit policies before enforcing them in production.
  • Track metrics like latency, error rates, and usage volume.

Once you validate a server, you can promote it to production with the same policies and MCP Gateway visibility.

note

The LLM Gateway bridges your sandbox and production environments ensuring consistency, security, and clear observability across all MCP traffic

Best practices

  • Always test external or community-hosted MCP servers in the sandbox before production.
  • Use mock or anonymized data during validation.
  • Apply short-lived credentials through secret management for testing.
  • Clean up expired sandbox instances regularly.
  • Use audit logs to document your validation process for compliance.
On this Page