Safeguard
Safeguard MCP Server

Security for AI Assistants

Universal Model Context Protocol server — bring vulnerability scanning, SBOM analysis, and automated remediation to Claude, ChatGPT, Grok, and your IDE.

Claude ConnectorHosted ServiceUptime SLA
◈ the switchboard · scoped, scrubbed, signed
89+
MCP tools for security operations
3
AI assistants supported out-of-the-box
~500ms
Typical response time (p95)
High
Production uptime, backed by SLA
The gateway

One Server, All Your AI Tools.

Universal AI Integration

01

Single API gateway for Claude Desktop, ChatGPT, Grok, VS Code, IntelliJ, and any MCP-compatible client.

Enterprise Security

02

OAuth 2.1 Device Flow + API Key authentication. Auto-token refresh, multi-tenant support, and complete audit trail.

89+ Security Tools

03

Comprehensive tools for vulnerability scanning, SBOM management, automated remediation, and compliance reporting.

Natural Language Interface

04

Ask questions in plain English. AI assistants understand your security queries and provide actionable insights.

Complete toolkit

Everything You Need for Security.

Vulnerability Scanning

Find CVEs by severity, CVSS, ecosystem

Automated Remediation

Fix npm, pip, maven, go, cargo dependencies

SBOM Management

Create, compare, analyze Software Bill of Materials

Security Metrics

Risk scores, compliance reports, trend analysis

SCM Integration

GitHub, GitLab, Bitbucket, Azure DevOps

Compliance

PCI-DSS, HIPAA, SOX, GDPR, FedRAMP templates

Natural Language

Ask questions and get AI-powered answers

CI/CD Ready

API key auth for automated workflows

Workflows

Built for Modern Workflows.

AI-Powered Security Reviews

Let Claude or ChatGPT analyze your projects for vulnerabilities and compliance issues

Automated Remediation

AI assistants automatically fix vulnerabilities in package.json, requirements.txt, and more

IDE Integration

VS Code and IntelliJ plugins connect to MCP server for real-time security insights

See It In Action

Ask in English. Get a verdict.

claude desktopvia mcp.safeguard.sh
User:
"Find all critical vulnerabilities in my project"
Claude:
Found 3 critical vulnerabilities:
• CVE-2024-1234 in express@4.17.1
• Fix available: Update to 4.18.2
vs code pluginclient.ts
const mcpClient = new SafeguardMcpClient(
  "https://mcp.safeguard.sh"
);

const result = await mcpClient.callTool(
  'safeguard_remediate_npm',
  { package_json_content }
);
quick connectclaude_desktop_config.json
# 1. Get your API key at app.safeguard.sh
Settings → API Keys → Generate API Key
# 2. Add to Claude Desktop config
{
  "mcpServers": {
    "safeguard": {
      "url": "https://mcp.safeguard.sh/mcp/sse",
      "transport": "sse",
      "env": {
        "SAFEGUARD_API_KEY": "sg_api_your_key"
      }
    }
  }
}
# 3. Restart Claude Desktop
✓ Ready! Ask Claude to use Safeguard tools
Core capabilities

Governance for every agent call.

Six controls that turn agentic AI from a liability into an auditable surface.

Tool allowlist + capability scoping

Declare which MCP tools an agent can call, with per-tool capability bounds. Restrict to read-only, sandboxed write, or fully privileged surfaces per agent.

Sensitive-data egress guardrails

Inline scanning of tool outputs for PII, secrets, and proprietary code before they leave the boundary. Redact, block, or quarantine in milliseconds.

Prompt audit log

Every prompt, tool call, and response is stored with cryptographic chain-of-custody. Replay any session for incident response or regulatory review.

Per-agent identity

Each agent and model gets its own credentials and quota. Revoke a single misbehaving client without disrupting the rest of the fleet.

Real-time policy enforcement

Block, redact, or require approval based on context — file path, repo, requester, or downstream destination. Decisions resolve in single-digit milliseconds, typically.

Drop-in for any MCP-aware client

Works with Claude Code, Cursor, Cline, and custom agents out of the box. Standards-compliant transport with SSE and HTTP support.

Use cases this product solves

Real deployments, real outcomes.

Lock down an agentic coding assistant in a financial-services repo

Setup

Point the assistant at the MCP server instead of direct API access.

Block writes to production manifests and redact customer IDs in tool output. The assistant can read code freely, but every mutation is gated by policy and every emission is scrubbed of identifiers.

Metric

Customer-ID egress blocked at the boundary in pilot

Comply with EU AI Act tier-2 logging in the SDLC

Setup

Route every AI-assisted change through the MCP boundary.

Capture prompts, tool calls, and responses with chain-of-custody — without writing your own audit infrastructure. Export to your evidence pipeline on a schedule or on demand for an auditor.

Metric

Full tier-2 evidence in days, not quarters

Prevent prompt-injection secret exfiltration

Setup

Wrap the agent's vector store and tool calls behind MCP.

When a poisoned document tries to coerce the agent into dumping a token, the egress scanner catches the secret pattern in the tool response before it reaches the model context. Quarantine and alert in one step.

Metric

Secret patterns caught before reaching model context

Govern multiple AI vendors through one allowlist

Setup

Front Claude, GPT, and in-house models with the same MCP server.

Manage policy once instead of in N vendor consoles. Add a new vendor by issuing an identity; remove one by revoking it. Audit logs unify across providers with a single schema.

Metric

One control plane for every AI client

Request lifecycle

End-to-end request lifecycle.

01

Client connects

An MCP-aware client opens a session against the Safeguard MCP endpoint.

02

Auth + identity check

Per-agent credentials are validated and bound to a tenant + policy set.

03

Tool capability scope applied

Allowlist filters the tools the agent may invoke and the arguments allowed.

04

Tool call inspected by policy engine

Each call is evaluated against context — repo, path, requester, time.

05

Output scanned for sensitive data

Secrets, PII, and proprietary code are redacted or blocked at egress.

06

Audit log entry

A signed record is written with prompt, call, response, and decision.

07

Response returned

The sanitized result is streamed back to the client in standard MCP shape.

Ready to Secure with AI?

Join developers using AI assistants for automated security and compliance.

Cloud HostedSOC 2 Type II (audit in progress)Free Tier Available

Self-healing security runs on Safeguard.

Your first fix PR is minutes away.

No sales call required, even your agent can complete the purchase over MCP.