MCP server

Sentrasec ships a Model Context Protocol server as part of the CLI. Any MCP-capable agent can use it to scan code, read findings, explain vulnerabilities and propose fixes.

Per-client setup guides live at mcp.sentrasec.ai.

How it runs

sentrasec mcp

The server speaks JSON-RPC over stdio as a subprocess of your agent. No network port is opened, and it acts as the authenticated user: an agent cannot read findings you cannot read.

Authenticate once before starting:

sentrasec login

Available tools

Tool	Purpose
security.dast_scan	Run a DAST scan against a URL and return enriched findings
security.list_findings	List findings for a scan or across the workspace
security.explain	Explain a vulnerability by knowledge-graph node id
security.suggest_fix	Suggest a code-level fix for a finding
security.list_rules	List detection rules available in the engine
security.list_languages	List supported languages
security.issue_cert	Issue a pentest certificate for a set of scans

Tools follow the standard MCP schema. Clients discover them on connect: you ask in natural language and the agent selects the tool.

Claude Code

claude mcp add sentrasec -- sentrasec mcp

Cursor

In .cursor/mcp.json:

{
  "mcpServers": {
    "sentrasec": {
      "command": "sentrasec",
      "args": ["mcp"]
    }
  }
}

VS Code

In .vscode/mcp.json:

{
  "servers": {
    "sentrasec": {
      "type": "stdio",
      "command": "sentrasec",
      "args": ["mcp"]
    }
  }
}

GitHub Copilot

In the agent MCP configuration:

{
  "mcp": {
    "sentrasec": {
      "command": "sentrasec",
      "args": ["mcp"]
    }
  }
}

Using it

Once connected, ask the agent in plain language:

Scan this file for security issues before I commit it.

Why is this finding a problem, and what should I change?

Is the version of this package we use affected by any known vulnerabilities?

The agent selects the appropriate tool, Sentrasec answers from the same engine and knowledge graph as every other surface, and the response arrives inline.

Scope and safety

Scans run locally: source does not leave your environment to be analysed. The server exposes read and analysis operations; it does not modify your code. Fix suggestions are returned to the agent as proposals for you to apply.