Scanning for secrets with the GitHub MCP server

Detect exposed secrets in real time from your AI coding agent, before they ever reach your repository.

Wer kann dieses Feature verwenden?

People with write access to a repository.

Tool navigation

In diesem Artikel

The GitHub Model Context Protocol (MCP) server lets you run secret scanning directly from GitHub Copilot agent mode, GitHub Copilot-CLI, and other MCP-compatible tools. Scan your code for exposed keys, tokens, and credentials as you work, and fix them before you push.

The secret scanning tools are only available via the GitHub remote MCP server. Local MCP server configurations are not supported.

This works with any MCP-compatible agent or IDE, including Visual Studio Code, JetBrains, Claude Code, Cursor, and Windsurf. The experience varies across clients.

Hinweis

Findings returned by MCP-invoked scans are ephemeral. They are surfaced in your agent's chat for the current session only and are not persisted as alerts on GitHub. This means these findings won't appear in the Security tab, in the secret scanning alerts list, or in the REST/GraphQL APIs for alerts. MCP scans should be treated as a pre-commit safety check, not as a system of record. Remediate findings before they are pushed to the repository and persisted in Git history.

Prerequisites

  • GitHub Secret Protection is enabled for the repository.
  • GitHub MCP server is connected in your IDE or agent. See Einrichten des GitHub MCP-Servers.
  • Your organization's security configuration determines which secret types are detected and whether push protection is enforced. The MCP tools respect your organization's push protection configuration (repository-level push protection settings are not used).

Step 1: Install and configure tools

Enable the secret_protection toolset

Enable the secret_protection toolset to make the scanning tools available to your agent. The default toolsets do not include it.

The run_secret_scanning tool is currently attached to the copilot toolset rather than secret_protection. You must explicitly include run_secret_scanning as an additional tool alongside the secret_protection toolset in your MCP configuration.

GitHub Copilot-CLI has the GitHub MCP server built in:

copilot mcp --toolsets=secret_protection --tools=run_secret_scanning

Add the secret_protection toolset and the run_secret_scanning tool to your MCP configuration:

JSON
{
  "servers": {
    "github": {
      "url": "https://api.githubcopilot.com/mcp/",
      "headers": {
        "X-MCP-Toolsets": "secret_protection",
        "X-MCP-Tools": "run_secret_scanning"
      }
    }
  }
}

In your JetBrains IDE, edit your MCP server configuration to include the secret_protection toolset and run_secret_scanning tool headers. For more information on configuring MCP servers in JetBrains, see MCP Server in the JetBrains documentation.

JSON
{
  "servers": {
    "github": {
      "type": "http",
      "url": "https://api.githubcopilot.com/mcp/",
      "headers": {
        "GitHub-MCP-Toolsets": "secret_protection",
        "GitHub-MCP-Tools": "run_secret_scanning"
      }
    }
  }
}

(Optional) Install the Advanced Security plugin

The Advanced Security plugin gives you a /secret-scanning slash command for a streamlined scanning experience in GitHub Copilot-CLI and Visual Studio Code. The plugin uses the MCP tools under the hood, so you'll still need to enable the secret_protection toolset.

Instructions for installing the plugin:

Step 2: Scan your code

Once the toolset is enabled, you can trigger a scan in several ways depending on your client.

Natural-language prompt. In any MCP-compatible agent, you can ask:

"Scan my current changes for exposed secrets and show me the files and lines I should update before I commit."

"Run secret scanning on the files I've changed since my last commit and summarize any high-confidence findings."

Slash command (requires the Advanced Security plugin). If you installed the optional plugin in Step 1, you can also use:

"/secret-scanning Review the staged diff for credentials, keys, or tokens and propose replacements using environment variables."

Direct tool invocation: You can also invoke the scanning tool directly from your client.

Run copilot --add-github-mcp-tool run_secret_scanning.

Type /secret-scanning in Copilot-Chat.

  1. In your IDE, open Copilot-Chat
  2. Click the Agent tab
  3. Use a prompt like: "Scan my recent changes for exposed secrets before I commit." You can also click the tools icon in the chat box to browse available secret_protection tools directly.

The agent returns:

  • The type of secret found
  • The file and line where it was detected
  • Remediation steps, such as removing or rotating the credential

If push protection is enabled, the MCP server also blocks secrets from being included in any actions it takes on your behalf, such as commits, pull requests, or the creation of files. See Arbeiten mit Pushschutz und dem GitHub MCP-Server.

Troubleshooting

ProblemCheck
Scanning returns no resultsVerify the secret_protection toolset is enabled in your MCP configuration.
Repository not eligibleSecret scanning via MCP is available to public repositories and requires GitHub Secret Protection to be enabled for private and internal repositories.
Agent doesn't recognize the toolConfirm your IDE or agent supports MCP. See Informationen zum Model Context Protocol (MCP).
Unexpected detection resultsYour organization's security configuration controls which patterns are scanned. Check your repository security settings.
Tool works in one client but not anotherThe experience varies across MCP-compatible clients. Check your client's MCP documentation for supported features.

Further reading