Gadriel AI Security Harness

Name: Gadriel AI Security Harness
Author: Gadriel-ai

Local

gadriel.aiSecurity0Updated 1d ago

Repository Visit website

About Gadriel AI Security Harness

Gadriel AI Security Harness scans code written by GitHub Copilot for conventional and AI-specific security risks. It runs locally and provides Copilot with MCP tools, repository guidance, prompts, and reviewer agents.

Key features

01Local SAST and secrets scanning
02Dependency, container, and configuration checks
03Detection of prompt injection and OWASP LLM risks
04MCP tools for findings and fixes
05Copilot repository instructions and reviewer agents

Use cases

01Scan a repository written with Copilot
02Inspect findings for a specific file
03Ask Copilot to fix a security finding

MCP support

The MCP server exposes scanning and finding-management tools to Copilot, including file and buffer validation, finding lookup, and fixes. Scans run locally; they are invoked through MCP tools and prompts rather than automatically on every edit.

Configuration

Local server config is the same for every client. Copy it into your client’s MCP config file.

mcp.json · Universal config

{
  "mcpServers": {
    "gadriel-ai-gadriel": {
      "command": "npx",
      "args": [
        "-y",
        "gadriel"
      ],
      "env": {
        "GADRIEL_NO_ANONYMOUS_AUTH": "<GADRIEL_NO_ANONYMOUS_AUTH>"
      }
    }
  }
}

Run commandnpx -y gadriel

Environment variables to set

GADRIEL_NO_ANONYMOUS_AUTHSecretSet to 1 to skip anonymous device registration with app.gadriel.ai.

Local servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.

Packages

npmgadriel1.7.0stdio

Tools

No tool list yet

Local servers expose tools only once running on your machine, and the maintainer has not provided a tool list.

Details

From the project README

Gadriel — AI Security Harness for GitHub Copilot

Security scanning for the code Copilot writes: SAST, secrets, dependencies (SCA/SBOM), containers and configuration, including AI-specific risks like prompt injection and the OWASP LLM Top 10 — 3,000+ rules, scanned on your machine. Gadriel plugs into Copilot as an MCP server plus repository instructions, prompts, and reviewer agents.

Part of the Gadriel AI Security Harness, alongside VS Code, Claude Code, Codex, and Cursor.

Get it

Easiest — the VS Code extension. Install Gadriel AI Security Harness from the Marketplace. It registers the gadriel MCP server for Copilot automatically and adds a Gadriel: Scan Repository command — no config to edit.

Or add the MCP server yourself. The server is published to the GitHub MCP Registry as io.github.Gadriel-ai/gadriel, so it shows up in VS Code's MCP: Browse Servers and Copilot Chat's @mcp search — add it in a click. To wire it per-repo instead, drop this .vscode/mcp.json into your project (needs Node for npx, or npm install -g gadriel):

{ "servers": { "gadriel": { "type": "stdio", "command": "npx", "args": ["-y", "[email protected]", "code", "mcp"] } } }

Add the Copilot guidance (optional). Copy the .github/ directory into your repo so Copilot knows how to use Gadriel:

Path	What
`.github/copilot-instructions.md`	repo-wide guidance, auto-applied
`.github/instructions/*.instructions.md`	17 topic rules, scoped by `applyTo`
`.github/prompts/*.prompt.md`	`/gadriel-scan`, `/gadriel-fix`, `/gadriel-status`, …
`.github/agents/*.agent.md`	8 reviewer agents

Use

In Copilot Chat (agent mode), just ask: "Run a Gadriel security scan on this repo and summarize the findings," or invoke a prompt like /gadriel-scan. The gadriel MCP tools — validate_file, findings_for_path, fix_finding, validate_buffer, and more — are available to Copilot directly.

Good to know

No automatic edit guardrail here. Copilot/VS Code has no edit-time hook (Cursor and Codex do), so scanning runs on request via the MCP tools and prompts rather than blocking each edit.
Enterprise: Copilot Business/Enterprise can allowlist the gadriel MCP server through managed MCP policy.
Privacy: code is scanned locally and never uploaded. First run registers an anonymous device credential with app.gadriel.ai (a random device id — no hostname, username, or keys); set GADRIEL_NO_ANONYMOUS_AUTH=1 to skip. See the privacy policy.

Maintainers

The registry listing is (re)published by .github/workflows/publish-mcp.yml (GitHub OIDC — an org namespace can only be published from CI in a Gadriel-ai repo). After a new gadriel npm release, bump server.json and the .vscode/mcp.json pin, then re-run that workflow.

License

This repository is Apache-2.0. The gadriel scanner it runs is proprietary, under the Gadriel terms.

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

Read the getting-started guide

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.