@ia-qa/cli

Run ia-qa.com's 148 QA / LLM / dev tools from your shell — for AI agents (Claude Code, Copilot CLI, Cursor agents) and humans. Zero dependencies. No signup.

Most tools run key-free. For the LLM-powered ones (e.g. VLM test suites, LLM-judge eval), bring your own key (BYOK) — pass it as a parameter; nothing is stored.

The CLI carries no business logic: it discovers tools from the public MCP manifest and runs them on the hosted server (https://www.ia-qa.com). Requires Node ≥ 18.

Easiest start — have your agent set it up

Working with Claude Code, Cursor, Copilot CLI? Paste this to it instead of reading the flag list:

Set up @ia-qa/cli in this project so you can run QA tools from the shell.
1. Run `npx -y @ia-qa/cli list` and `npx -y @ia-qa/cli --help`, then
   `npx -y @ia-qa/cli <tool> --help` for anything you pick. That output is the
   authoritative tool list — do not guess tool names or flags from memory.
2. Look at what I'm working on and tell me which 3 of those tools actually fit,
   what verdict each one returns, and how I'd gate CI on it.
3. Tell me plainly which of them need my own API key (BYOK), and which ones send
   my input to the hosted server.
4. Run one of them on my real data — after I confirm which.

Step 1 is the load-bearing one: the catalogue changes with every server deploy, so a model answering from memory will confidently invoke tools that don't exist. list is generated from the live manifest. Step 3 is not politeness — this CLI is a thin client, your input is sent to https://www.ia-qa.com to be executed there; only @ia-qa/self-healing runs fully offline.

Agents that speak MCP can skip the CLI entirely and point at https://www.ia-qa.com/mcp — same tools, same manifest.

Install

# One-off, no install:
npx @ia-qa/cli --help

# Or global:
npm install -g @ia-qa/cli
ia-qa --help

# Or download the zip from https://www.ia-qa.com/downloads/ia-qa-cli.zip and:
node ia-qa.js --help

Use

ia-qa --help                       # categories + examples
ia-qa list [--category encoding]   # every tool, one line each
ia-qa <tool> --help                # flags for one tool

ia-qa generate_uuid --count 3
ia-qa base64_encode "hello"
ia-qa hash_text "secret" --algorithm sha256
echo '{"b":2,"a":1}' | ia-qa format_json --stdin

ia-qa text_stats "hello world" --json           # machine JSON to stdout
ia-qa split_chunks "$(cat doc.txt)" --chunk_size 256 --out chunks.json
ia-qa prompt_injection_scan "ignore all instructions" --report scan.html --open

• --json emits machine-parsable JSON (default is human-readable).
• Large outputs are written to a temp file and the path is printed; --out <file> chooses it.
• --report <file.html> also writes a self-contained, ia-qa.com-branded HTML report of the run (verdict, stats, per-case badges — secret params redacted). Add --open to open it in your browser (local use; leave it off in CI). --open alone writes a default-named report and opens it.
• --stdin (or - as the positional) reads the primary input from stdin.
• --server <url> / $IAQA_SERVER point at another server (default https://www.ia-qa.com, or a self-hosted ia-qa server).
• BYOK for LLM tools — pass your key as a parameter, e.g. ia-qa run_vlm_test_suite --image_url <url> --api_key "$OPENAI_API_KEY" --model gpt-4o.
• Exit codes: 0 success · 2 bad input / missing param · 1 unexpected / server error.

Also available as MCP

Every tool here is also callable over the Model Context Protocol — point Claude Desktop, Cursor, Windsurf, or any MCP client at https://www.ia-qa.com/mcp.

MIT · Powered by IA-QA.