Verified

Confirm MCP Server

Human-in-the-loop approval requests for sensitive agent actions using confirm.dev.

Data last scanned today · Reviewed today

352 tools★ 2TypeScriptMIT

Overview

Confirm MCP Server provides a structured toolset to pause AI agent execution for human oversight. By integrating the request_approval tool, agents can present specific actions, payload data, and reasoning to a human reviewer at confirm.dev before proceeding with sensitive operations. This creates an enforcement layer that operates independently of the agent's internal logic. The server includes a check_approval tool to facilitate polling on pending requests, ensuring that agents can wait for a human decision or proceed once a status change occurs. Every approval decision is stored in an immutable audit log, allowing for transparency and verification of actions performed by the agent.

Our verdict

This server carries a high risk level with a security score of 30/100. The rating reflects significant trust concerns, as the project is community-maintained rather than official and lacks editorial review. Given the sensitive nature of authorizing agent actions through an external service, potential users should be aware that the authentication and read-only mechanisms have not yet been evaluated for security. It is suitable for experimental environments or non-critical prototyping where manual intervention is required, but it is not currently recommended for sensitive production workflows.

Provides an immutable audit log for all agent decisions.
Allows human reviewers to edit payloads before final approval.
Small supply-chain footprint with only three direct dependencies.
Supports configurable wait times and TTL for approval requests.

High-risk security score due to lack of independent review.
Single contributor increases long-term bus-factor risk.
No official vendor maintenance or established security vetting.
Authentication and access controls remain unreviewed.

Setup

Claude Desktop

{
  "mcpServers": {
    "confirm-mcp": {
      "args": [
        "-y",
        "confirm-mcp"
      ],
      "command": "npx"
    }
  }
}

Paste into ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows), then restart Claude Desktop. Some servers require API keys or tokens as environment variables — check the project's README and add an "env" object if needed.

No read-only mode — running this grants full read/write access.
No built-in authentication — anyone with access to this config can invoke every tool it exposes.

Tools

Tool	Description	Risk
request_approval	Pause and ask a human to approve a sensitive action.	medium
check_approval	Poll a previously created request by id to see whether a human has decided.	low

Compatibility

Client	Local	Docker	Remote	Read-only
ChatGPT
Claude Desktop
Cursor
VS Code
Windsurf

Frequently asked questions

›What are the prerequisite requirements to run this server?

You must have Node 18 or higher installed and a valid API key obtained from confirm.dev.

›How can I notify a specific team of approvers?

You can use the notify parameter with the group:<key> format to escalate a request to an approver group.

›What happens if a human edits the action during the approval process?

If approved, the tool returns the effectivePayload, which incorporates any edits made by the human, ensuring the agent uses the modified data.

›Can the agent check the status of a request later if it does not receive an immediate response?

Yes, if the wait_minutes elapses while the request is still pending, the tool returns a request ID that the agent can use to poll for updates via the check_approval tool.

›Is it recommended to rely solely on this tool for critical production workflows?

For production environments requiring guaranteed coverage, the documentation recommends using the deterministic policy engine found in the @confirm/sdk instead of relying on the model to call the tool.