Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more

You can require a person to approve MCP tool calls without changing the MCP server only if the host application already offers an approval control. MCP itself has no universal setting that forces every client to show a confirmation prompt. Check your host’s controls first; if it has no blocking approval option, a genuinely human-gated call requires a different host or an application-side integration.

Why MCP has no universal approval switch

MCP defines how clients discover and invoke tools, including the tools/call request. It leaves the interaction pattern to the application, so one client’s approval setting does not automatically apply to another. The Model Context Protocol tools specification recommends that applications show exposed tools, indicate when they are invoked, and provide confirmation prompts. It says there “SHOULD always be a human in the loop with the ability to deny tool invocations,” but that is not a portable configuration field that makes every client pause.

Tool descriptions and annotations do not enforce approval either. The specification says clients must treat annotations as untrusted unless they come from a trusted server. The checkpoint must be implemented by the host—or by a trusted layer that can stop dispatch—rather than assumed from server metadata.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

Find out whether your host blocks each call

Before changing settings, identify the MCP host and the path in use: a local server, a hosted tool, or an agent framework may expose different controls. Check whether the approval applies to an individual call, a tool, a server, or a broader workspace; whether it blocks execution until a person responds; and what happens after denial or timeout. A trust prompt shown once is not necessarily approval of every future call.

Enable approvals in supported hosts

VS Code

  1. Open the Command Palette and run Chat: Manage Tool Approval.
  2. Find the MCP server under its source and configure either the source or the individual tool.
  3. To require manual approval rather than allow auto-approval, set the tool to false in chat.tools.eligibleForAutoApproval, as described in the VS Code tool approval guide.

The guide also documents organization device-management policies for specific tools. Names and available controls can vary by installed VS Code version, so check the setting in your current build.

Claude Managed Agents

In Claude Managed Agents, MCP toolsets default to always_ask. That policy pauses the session for approval; it can be configured for a toolset or overridden for an individual tool. The host handles the confirmation event and sends an allow or deny decision. These are Managed Agents policies, not a general setting for Claude Desktop, claude.ai, or every MCP client. See the Managed Agents tool permissions documentation.

Do not confuse auto with a required human checkpoint. Under auto, the server may allow, deny, or pause a call based on its evaluation, and safe calls may proceed without a person. The documentation explicitly says auto is not a human checkpoint; use always_ask for calls that must wait for a person.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

OpenAI Agents SDK

For hosted MCP tools, the OpenAI Agents SDK documents requireApproval: 'always' or a per-tool approval policy. A call that needs approval becomes an interruption, and the application resumes the run after a person approves or rejects it. The documented example implements a confirmation handler and responds to interruptions, so enabling the option is not a point-and-click change for an existing deployment. The option reference lists never as the default; verify the SDK version and deployed agent configuration. See the Agents SDK MCP guide and MCP option reference.

ChatGPT and Codex configurations

The reviewed ChatGPT connectors documentation says Codex MCP configuration is stored in config.toml and shared by the desktop app, CLI, and IDE extension for the same Codex host. It distinguishes that local Codex configuration from hosted ChatGPT tools. This establishes where some configuration lives; it does not establish a universal setting that prompts before every MCP call. Look for approval controls in the exact host and version you use rather than assuming a server configuration file can enforce a client-side pause.

If your host cannot require approval

If the existing host has no control that blocks execution until a person responds, there is no server annotation or universal MCP toggle that can add one. Your practical options are to use a host with a suitable approval UI or policy, or add a blocking approval step in the application or orchestration layer. The latter requires an integration that pauses dispatch, collects a decision, and then allows or rejects the call; the Agents SDK interruption flow is one documented example.

Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Support on Ko-Fi

Review each approval request carefully

An approval prompt is useful only if you can tell what will run. VS Code recommends reviewing tool names and input parameters, especially for actions that modify files, run commands, or access external services. Connector guidance also warns that tool behavior can change and that malicious instructions may be hidden in content. Treat external tool output as untrusted, and do not grant persistent trust just to dismiss prompts.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.
  • Check the server and tool identity, then inspect the arguments for the specific call.
  • Approve only the action you intend; deny calls whose scope or effect is unclear.
  • Keep approval scoped to an individual tool or call where possible, and use persistent allow choices sparingly.
  • Disable tools you do not need.

Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.