Customer Guidance

Security Best Practices

Use this page to safely connect Smplify MCP in Claude Code, Claude Desktop, OpenAI Codex, Cursor, VS Code Agent Mode, and other trusted clients that support remote Streamable HTTP MCP with OAuth. This guidance complements the live security overview documented on /security.

Black and white illustration representing a completed security review and operational checklist

Protected access, human oversight, and safer operational boundaries for your connected MCP workflows.

Only connect to the official Smplify MCP endpoint: https://mcp.smplify.com/mcp. Hosted configurations must not contain an upstream API key, bearer header, or token in the URL.

Use trusted clients and installers

Only add Smplify MCP in clients and marketplaces that you and your organization trust.

  • Verify the exact domain before installing: mcp.smplify.com.
  • Prefer official Smplify docs and the generated setup snippets for Claude, Codex, Cursor, and VS Code.
  • Be cautious with third-party “one-click install” marketplaces and copied shell commands.

Understand OAuth scope

Smplify MCP acts with the scopes you authorize and the permissions of the connected Smplify account.

  • Request only the MCP scopes needed for the client’s workflows.
  • Revoke authorizations that are no longer used or may have been exposed.
  • Let the client complete OAuth discovery; never paste an upstream Smplify API credential into a hosted MCP configuration.

Watch for prompt injection

Agents can be influenced by malicious content in tickets, notes, dashboards, emails, or documents connected to the same workflow.

  • Do not let external content automatically override your operating instructions.
  • Treat requests to export data, disable safeguards, or ignore policy as suspicious.
  • Review destructive or cross-system actions carefully before approval.

Keep a human in the loop

Human confirmation is strongly recommended for administrative actions that affect devices, users, or policies.

  • Require approval for device lock, wipe, unenroll, or authentication changes.
  • Require approval for tenant, directory (Entra/Google/LDAP), identity/role, enrollment, software-update, and policy updates.
  • For exploratory work, prefer review-first or read-only workflows.

Official Smplify domains

  • https://mcp.smplify.com/mcp — official Smplify MCP endpoint
  • https://mcp.smplify.com/setup — official setup guidance and generated client config snippets
  • https://mcp.smplify.com/security — security overview and control boundaries
  • https://api.smplify.com/api/v1 — upstream Smplify API base URL

Recommended customer practices

  • Keep URL-based token auth disabled and reject upstream bearer-token passthrough at the hosted endpoint.
  • Use least-privilege OAuth scopes and Smplify account roles.
  • Document who can approve destructive actions and who owns client authorization and revocation.
  • Re-test OAuth discovery, login, refresh, and reconnect flows after configuration changes.
  • Use trusted clients and leave hosted configurations headerless so the client follows OAuth discovery.
Black and white illustration representing explicit approval and completion of an administrative task

Use explicit approval steps for device, policy, role, and tenant actions.

Black and white illustration representing a security checklist and operational review

Keep a repeatable checklist for authorization review, audit verification, and workflow review.

What Smplify operates and enforces

Smplify operates both the hosted MCP gateway and the underlying Smplify Platform APIs. Requests run inside Smplify-managed controls using a Smplify-issued MCP token and the tenant permissions resolved by the authorized connector.

  • The setup portal never asks for or handles a Smplify username, password, API key, or bearer token.
  • The hosted service accepts only Smplify-issued MCP tokens bound to https://mcp.smplify.com/mcp.
  • Smplify MCP does not override the Smplify account permissions resolved by the authorized connector.
  • If you connect additional third-party tools to the same agent, those tools keep their own permissions and data-handling boundaries.