Skip to main content

​
Prerequisites

  • Python 3.11 or later
  • uv (recommended) or pip
  • Node.js / npm — MCP server targets are typically npm packages. Install the example target used below with npm install -g @modelcontextprotocol/server-memory.
  • LLM provider API key (optional) — Required only for inject campaigns and chain execution. q-ai uses litellm for multi-provider support. Model strings use the provider/model format (e.g., anthropic/claude-sonnet-4-20250514, openai/gpt-4o, groq/llama-3.3-70b-versatile). Set the matching environment variable (e.g., ANTHROPIC_API_KEY, OPENAI_API_KEY, GROQ_API_KEY) or store credentials in the OS keyring via qai config set-credential.
  • sentence-transformers + chromadb (optional) — Required only for rxp module. Install with pip install q-uestionable-ai[rxp].

​
Install

pip install q-uestionable-ai

​
Configure a provider (optional)

If you plan to run injection campaigns, store your API key in the OS keyring: 
qai config set-credential anthropic
# Prompts for your API key (masked input)
Verify configured providers: 
qai config list-providers
Alternatively, set the environment variable directly: 
export ANTHROPIC_API_KEY="sk-ant-..."

​
Run your first scan

1

Pick a target server

You can scan any MCP server you have permission to test. This guide uses the official MCP memory server as an example. You don’t need to start the server manually — q-ai spawns it via the --command flag.Install the example target if you don’t have one:
npm install -g @modelcontextprotocol/server-memory
2

Run the scan

q-ai launches the server as a child process, connects over stdio, enumerates its tools and resources, then runs all 10 OWASP MCP Top 10 scanner modules against it.
qai audit scan \
  --transport stdio \
  --command "npx @modelcontextprotocol/server-memory"
3

Review the output

The CLI prints a live summary as it runs:
  • Connected — confirms the transport and server name
  • Findings — any security issues found, with severity, description, and remediation
  • Report saved — path to the full JSON report

​
Enumerate a server

Before a full scan, you can quickly check what a server exposes: 
qai audit enumerate \
  --transport stdio \
  --command "npx @modelcontextprotocol/server-memory"
This lists the server’s tools, resources, and prompts without running any security checks.

​
Launch the web UI

Running qai with no subcommand starts the web UI: 
qai
This opens a browser with the workflow launcher — workflow cards for guided security assessments. The same findings are accessible via both the web UI and CLI (qai findings list). Use --port to specify a port, or --no-browser to start the server without opening a browser.

​
Try other modules

# Generate indirect prompt injection payloads
qai ipi generate --callback http://localhost:8080 --technique all --format markdown

# Launch the CXP interactive builder
qai cxp generate

# Validate RAG retrieval rank
qai rxp validate --profile hr-policy --model minilm-l6

# View all findings across modules
qai findings list

​
Next steps

Only test systems you own, control, or have explicit written authorization to test.