Installation

Last updated: Oct 28, 2026

This guide tells you how to install the Fluid Attacks MCP server in Claude Desktop, Claude Code, Cursor, VS Code, and Windsurf. Select your AI assistant in the tabs below to see its configuration steps and a usage example. Also, you can learn how to integrate the Fluid Attacks security scanners into your software development lifecycle (SDLC) with AI agents.

Prerequisites

• To use the Fluid Attacks MCP Server, you must have an AI assistant that supports MCP, for example, Claude (in Cursor, Claude Desktop, VS Code, or other tools).
• You must authenticate to use most MCP tools. Five documentation and scanner tools work without authentication. Claude Desktop, Claude Code, VS Code, and Windsurf use an OAuth flow with the Fluid Attacks platform for this. For Cursor, you must give a Fluid Attacks API token.
• To run security scans, you must have Docker installed. Also, Fluid Attacks recommends that you configure an AGENTS.md file in the root directory of the project.

Configure your MCP client

Select your AI assistant in the tab list below. Each tab has the configuration steps and a usage example for that client.

Claude DesktopClaude CodeCursorVS CodeWindsurf

Install the runtime environment

To use the MCP server of Fluid Attacks, you must have Node.js. Do these steps to install it. If you already have Node.js v22, go to the next section.

1. Fluid Attacks recommends that you install Node Version Manager (nvm). With nvm, you can have different versions of Node.js and change from one version to another when necessary. To install nvm, run this command in your terminal:

curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.40.2/install.sh | bash

Close your terminal and open it again. As an alternative, run this command to load nvm into your current session:

\. "$HOME/.nvm/nvm.sh"

2. Install Node.js version 22 with this command (the minimum version is v18):

nvm install 22

Set up in Claude

To configure the MCP server of Fluid Attacks in Claude, follow these steps:

1. In the menu bar, click Claude and then Settings.

2. Go to the Developer tab.
3. Click the Edit Config button.

4. A folder opens. This folder has the file claude_desktop_config.json. Open this file with a text editor.

5. Paste this text in the configuration file:

   {
     "mcpServers": {
       "fluidattacks-mcp": {
         "command": "npx",
         "args": [
           "-y",
           "mcp-remote",
           "https://app.fluidattacks.com/mcp/messages/"
         ]
       }
     }
   }

6. Save the file.

7. Close the file and Claude.

8. Open Claude Desktop again. When Claude Desktop starts, mcp-remote opens your default browser and shows the Fluid Attacks sign-in page. On this page, you give Claude Desktop access to the information in your Fluid Attacks platform account. After you authorize the request, the browser opens Claude Desktop again, and the fluidattacks-mcp server shows as connected.

Then, you can chat with the Fluid Attacks platform.

Usage example

After you connect the Fluid Attacks MCP server to your AI assistant, you can request vulnerability information or security scans in natural language. The same prompts work in Claude Desktop, Claude Code, Cursor, VS Code, and Windsurf.

For example, to generate a vulnerability report for a group in an organization, use this prompt:

Using Fluid Attacks' API, please generate a one-page report of the vulnerabilities found in the Narrabri group of the Imamura organization.

To find the vulnerability with the highest severity in a file, use this prompt:

From the Fluid Attacks API, what is the highest severity (CVSS) vulnerability in the basketitems.ts file of the Clickable group within the Imamura organization?

Your AI assistant asks for permission before it runs each tool of the MCP server. One of these tools is fetch_weakness_vulnerabilities. It gets vulnerability information, for example, the status and severity. After you give permission, the assistant shows the information that you requested. It can also help you remediate the vulnerabilities directly in your codebase.

Sandbox configuration

Sometimes, you cannot use the interactive OAuth flow. For example, you run the MCP server in a sandbox or headless environment, you call it from CI, or you write scripts for integration tests. In these conditions, you can authenticate with a Fluid Attacks API token in the Authorization header.

Add the headers field to the server entry in your client configuration. Then, set Authorization to Bearer <YOUR_API_TOKEN>.

The example below uses the .mcp.json format of Claude Code. For VS Code, add the headers field to the entry in the servers section.

{
  "mcpServers": {
    "fluidattacks-mcp": {
      "type": "http",
      "url": "https://app.fluidattacks.com/mcp/messages/",
      "headers": {
        "Authorization": "Bearer <YOUR_API_TOKEN>"
      }
    }
  }
}

For Windsurf, use the serverUrl field with the headers field in ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "fluidattacks-mcp": {
      "serverUrl": "https://app.fluidattacks.com/mcp/messages/",
      "headers": {
        "Authorization": "Bearer <YOUR_API_TOKEN>"
      }
    }
  }
}

Claude Desktop does not support the type and headers fields above. claude_desktop_config.json accepts only local (stdio) servers. Thus, give the token to the mcp-remote command with a --header argument:

{
  "mcpServers": {
    "fluidattacks-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://app.fluidattacks.com/mcp/messages/",
        "--header",
        "Authorization:Bearer <YOUR_API_TOKEN>"
      ]
    }
  }
}

Security considerations

• Data sensitivity: Be careful when you share vulnerability data.

Integrate our scanners into your SDLC using AI

You can configure AI development tools to run the Fluid Attacks security scanners automatically at the correct points in your SDLC. Examples are Cursor, GitHub Copilot, and other AI coding assistants. This integration helps you find and remediate security vulnerabilities as soon as possible.

Before you integrate the security scanners of Fluid Attacks with AI agents, make sure that you have the prerequisites. You must have Docker installed and the MCP server of Fluid Attacks configured in your AI development environment. Also, you must have write access to your project directory for the configuration files and scan results.

Add AGENTS.md to your project using Fluid Attacks' MCP server

In the chat of your AI agent, type /fluidattacks-mcp/. Then, select this command:

/fluidattacks-mcp/config_agents_md

Add AGENTS.md to your project manually

1. Create the AGENTS.md file

Add an AGENTS.md file to the root directory of your project. This file has the instructions that AI agents use to know when and how to run the security scanners.

The AGENTS.md file must be at the root of your project:

your-project/
├── AGENTS.md     # Add this file
├── src/
├── package.json
└── ...

2. Add the AGENTS.md content

# Project Instructions

## Execute Security Scanners

### Execute SCA Scanner When:

- New dependencies are added to the project
- Dependencies are updated to new versions
- Lock files are modified (e.g.,`package-lock.json`, `uv.lock`)
- User explicitly requests a dependency security scan
- Setting up a new project for the first time
- Before deploying to production
- Before pushing to the repository

### Execute SAST Scanner When:

- Source code changes are made to application files
- New features or modules are added
- Security-sensitive code is modified (authentication, authorization)
- User explicitly requests a code security scan
- Before committing significant code changes
- During code reviews
- Before deploying to production
- Before pushing to the repository

### Execute Both Scanners When:

- A complete security audit is needed
- Major project updates involving both code and dependencies
- Pre-deployment security check
- User requests a full security scan

## Prerequisites

- Docker installed on the system
- No Dockerfile creation needed - only download the Docker images
- Write access to the project directory for configuration files and results

## Fluid Attacks Scanner

### Purpose

Scan the project for vulnerabilities using the Fluid Attacks MCP tools.

### Step-by-Step Instructions

#### 1. Use Fluid Attacks MCP tools to configure and run the scanner

#### 2. Add the output file to .gitignore

#### 3. Remediate vulnerabilities

- Review the output file
- If there are vulnerabilities, remediate them

## Best Practices for Agents

### 1. Configuration File Management

- Always verify the correct paths for include/exclude before running
- Adjust configuration based on project structure
- Use `.gitignore` as a reference for exclude patterns
- Store configuration files in the project root and add them to .gitignore

## When to Run What

| Scenario                      | Scanner | Priority |
| ----------------------------- | ------- | -------- |
| New dependency added          | SCA     | High     |
| Code changes in auth/security | SAST    | Critical |
| Weekly security audit         | Both    | Medium   |
| Pre-deployment check          | Both    | Critical |
| Dependency version update     | SCA     | High     |
| New feature development       | SAST    | Medium   |
| Third-party library added     | SCA     | High     |
| API endpoint changes          | SAST    | High     |

## Integration with Development Workflow

- On Code Changes: Run SAST if source files modified
- On Dependency Changes: Run SCA if dependency files modified
- On User Request: Run appropriate scanner(s)
- Help with remediation: Always create/update security reports
- Re-scan: After fixes to verify remediation

Troubleshooting

If the scanners do not run automatically, do these checks:

• Make sure that AGENTS.md is in the root of your project.
• Make sure that the Fluid Attacks MCP server is configured correctly.
• Make sure that Docker is installed and running.
• Make sure that the AI agent has read access to AGENTS.md.
• Ask the AI agent to run the scanners: "Run fluid attacks scanners."

If the scanners report false positives, do these actions:

• Examine the details of the vulnerability.
• If necessary, change the scanner configuration.
• Write the accepted exceptions in your project documentation.
• If you have questions, contact the Fluid Attacks support team.