MCP Agent Graph Auditor | Multi-agent security

MCP guide · Agent Graph Auditor

MCP Agent Graph Auditor: multi-agent transitive agency — @guardbee/mcp-agent-graph-auditor

@guardbee/mcp-agent-graph-auditor does not stop at one agent’s tool list — it builds a reachability graph across multi-agent orchestration. An agent with no dangerous tool of its own that can still reach a shell tool via delegation or group membership is reported as critical.

What does it do?

It extracts agent/tool edges from LangGraph (structural add_node/add_edge), CrewAI (allow_delegation), and AutoGen/ag2 (GroupChat, register_function); classifies tools with a capability catalog; and BFS-finds both direct and transitive excessive agency.

Who is it for?

Multi-agent teams on LangGraph, CrewAI, or AutoGen
DevSecOps auditing agent orchestration on every PR
Developers who want this check inside Claude or Cursor

What it is not

A single MCP server’s tool definitions (use Server Auditor)
Cross-file Python reference resolution (single-file scope)
JS/TS LangGraph APIs (v1 is Python-only)

Features

Feature	Description
3 frameworks	LangGraph structural; CrewAI/AutoGen heuristic
Direct vs transitive	Delegation-crossing paths always critical
7 capability rules	code/process/fs-write/network/credentials
SARIF 2.1.0	CI/CD
guardbee.yml	fail-on, max-files, exclude

MCP tools

Tool	Description
`scan_text`	Scan a Python snippet
`scan_file`	Scan a single .py file
`scan_directory`	Scan an agents source directory
`list_patterns`	List the capability catalog

Install with npm

Global install (handy for Claude Desktop / Cursor):

bash

npm install -g @guardbee/mcp-agent-graph-auditor

or via npx (recommended with -y in config):

bash

npx -y @guardbee/mcp-agent-graph-auditor

npm: @guardbee/mcp-agent-graph-auditor

CLI: npx @guardbee/mcp-agent-graph-auditor scan ./agents --fail-on=high --format=sarif > results.sarif. Opt out of telemetry with GUARDBEE_TELEMETRY=0.

Claude Desktop configuration

Config file paths:

macOS~/Library/Application Support/Claude/claude_desktop_config.json
Windows%APPDATA%\Claude\claude_desktop_config.json
Linux~/.config/Claude/claude_desktop_config.json

Merge this block, then restart Claude Desktop:

claude_desktop_config.json

{
  "mcpServers": {
    "guardbee-agent-graph-auditor": {
      "command": "npx",
      "args": [
        "-y",
        "@guardbee/mcp-agent-graph-auditor"
      ]
    }
  }
}

Cursor and other MCP clients

In Cursor, add the same command / args / env under Settings → MCP. stdio MCP servers use the same contract as Claude Desktop.

Example prompts in Claude

“Is there transitive excessive agency in this agents folder?”
“Can the CrewAI researcher reach a shell tool?”
“Which capability rules do you support?”

See surface risk with Guardbee

MCP servers speed up security work in chat. Start a 14-day free trial for brand scans and BeeAI.

Start free MCP servers

Frequently asked questions

Vs AI Code Scanner?

AI Code Scanner looks at one agent/tool list; Agent Graph Auditor graphs delegation hops.

CrewAI false positives?

Heuristic — allow_delegation + shared Crew membership is a conservative assumption; review findings.

API key?

No — local Python source scanning.

MCP Agent Graph Auditor: multi-agent transitive agency