# Security Recipes [Health: Active]

**Category:** 🔒 Security  
**Repository:** https://github.com/stevologic/security-recipes.ai  
**GitHub Stars:** 2  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/security-recipes

## Description
Read-only CVE intelligence, remediation playbooks, and agent setup guides. Not a scanner.

## Claude Desktop Quick Installation
No generated config for this server: it installs through its own setup steps.

The project's own setup steps (from https://github.com/stevologic/security-recipes.ai), copied verbatim. Run them in order; they are not an mcpServers config.

1. Install:

```sh
docker build -f Dockerfile.mcp-server -t security-recipes-mcp .
```

2. Start the server:

```sh
docker run --rm -p 8123:80 security-recipes-mcp
```

3. Run:

```sh
python -m venv .venv
```

4. Install:

```sh
pip install -r requirements-mcp-server.txt
```

5. Start the server:

```sh
python mcp_server.py
```

6. Configure:

```sh
.\.venv\Scripts\Activate.ps1
```

## Documentation

## What Security Recipes MCP server does

Security Recipes exposes security-recipes.ai content through MCP so compatible AI agents can search remediation recipes and retrieve approved upstream MCP context. Its surrounding site contains a rolling catalog of Medium, High, and Critical CVEs, evidence-qualified canonical vulnerability records, remediation playbooks, agent setup guidance, and security workflow material.

The server is intended for read-only research and planning. It helps an agent connect an existing security finding with relevant evidence, remediation guidance, implementation prompts, required proof, rollback considerations, and human-review boundaries. It does not scan repositories or infrastructure, open tickets, deploy changes, operate a SOAR workflow, or provide production authority.

## How it works

The MCP endpoint makes selected Security Recipes context available to an MCP client. Recipe search can help an agent identify guidance for dependency, SAST, sensitive-data, base-image, CVE, and default-hardening remediation workflows. Approved upstream MCP context can add advisory, scanner, repository, or runbook information while keeping the Security Recipes layer read-only.

The underlying CVE catalog is assembled from integrity-verified NVD JSON 2.0 feeds, CISA KEV metadata, and vetted remediation patterns. Canonical records use evidence-qualified information, including affected-version details and a remediation authority. The site's seven-phase change contract covers discovery, assessment, mitigation, remediation, verification, rollback, and triage; it defines evidence, outputs, approval boundaries, and failure behavior rather than guessing a patch or fixed version.

Reach for the server after another security tool has produced a finding and an agent needs context to assess or plan remediation. Keep a human reviewer and the relevant execution controls outside the MCP server.

## Setup and configuration

Connect an MCP-compatible client to the Security Recipes MCP endpoint at `https://security-recipes.ai/mcp`. The project includes setup guidance for Codex, Claude Code, Cursor, GitHub Copilot, Devin, Shiba Studio, Hermes Desktop, and OpenClaw; the compatible client configuration depends on the client being used.

The integration is read-only, and the project describes it as optional. No environment variables or local deployment settings are required by the documented endpoint usage. Existing security tools should remain responsible for producing findings, while the agent uses this service for research and bounded remediation context.

## Tools and capabilities

- Search security remediation recipes and CVE-related guidance.
- Retrieve approved upstream MCP context for advisory, scanner, repository, or runbook workflows.
- Provide sourced CVE facts and affected-version evidence from the site's security intelligence catalog.
- Supply bounded remediation planning context with verification, rollback, approval, and stop-condition guidance.
- Support agent setup patterns for Codex, Claude Code, Cursor, GitHub Copilot, Devin, Shiba Studio, Hermes Desktop, and OpenClaw.

## Limitations and notes

This is a read-only context service, not an execution system. It does not scan code, dependencies, hosts, or other infrastructure; findings must come from existing security tools. It is also not a ticketing system, SOAR platform, deployment tool, or custom security toolkit.

Remediation guidance is designed to keep agent work bounded and reviewable. Plans can identify likely targets, evidence requirements, outputs, approval boundaries, rollback paths, and failure behavior, but they do not authorize production changes. Human review remains part of the intended workflow, and deployment authority should stay with the surrounding tools and processes.

_Full upstream README: https://allmcps.com/mcp/security-recipes/readme_