Back to Home
Liquescent-Development icon

mcprobe

by Liquescent-Development

Overview

A testing framework that validates Model Context Protocol (MCP) servers by simulating LLM agents interacting with them via synthetic users and LLM judges.

Installation

Run Command
mcprobe run examples/scenarios/simple_greeting.yaml --model llama3.2 --base-url http://localhost:11434

Environment Variables

  • GOOGLE_API_KEY
  • MCP_URL
  • MCP_TOKEN

Security Notes

CRITICAL: The `extract_tools_from_server` function in `src/mcprobe/generator/mcp_client.py` directly executes a user-provided `server_command` via `subprocess.run`. This is a severe command injection vulnerability, allowing arbitrary system commands to be run if the input to `--server` CLI option is untrusted. Similarly, `load_agent_factory` in `src/mcprobe/agents/adk.py` uses `importlib.util.spec_from_file_location` and `spec.loader.exec_module`, which can execute arbitrary Python code from a user-specified path (via `--agent-factory` CLI option). This represents a significant arbitrary code execution risk if inputs are not fully trusted.

Similar Servers

Stats

Interest Score0
Security Score3
Cost ClassHigh
Avg Tokens750
Stars0
Forks0
Last Update2026-01-17

Tags

mcptesting-frameworkllm-agentsevaluationsynthetic-users