Argus

An MCP server that tests apps like a real testing engineer—exploring user journeys, discovering unscripted bugs, and proving each finding before reporting it.

Argus is an MCP server. It adds evidence-first browser QA to Claude Code, Codex, Cursor, or any MCP host without taking over the host agent's identity or broader coding task. The agent explores, inspects, verifies persistence, and records reproducible bugs. Every certified finding is independently re-confirmed from a clean page load before it's reported.

Product page · Quick start · Why Argus · Compared · Tools · Benchmarks

The output

Give it a URL; get a report of bugs — each tagged with whether Argus independently reproduced it or only observed it:

The green badge is the whole point. Anyone can have an LLM claim a bug. Argus re-loads the page from scratch and re-checks the symptom before it says VERIFIED — so the report is a list of bugs you can trust, not a list of guesses to triage.

Related MCP server: Aiqaramba

How it works

flowchart LR
    A(["observe"]) --> B{"looks wrong?"}
    B -->|not sure| C["act: click · type · resize · verify"]
    C --> A
    B -->|bug| D["verify_persistence — reload from a clean state"]
    D -->|symptom repeats| E(["VERIFIED"])
    D -->|symptom gone| F(["dropped — no false positive"])
    E --> G[["report: HTML · JSON · JUnit · SARIF"]]

The agent is the intelligence. Argus supplies concise QA guidance, a description-keyed tool surface (click_what("Login button"), not click(7)), a goal coverage ledger, and a reproduction-receipt engine that turns "the model thinks this is a bug" into "this bug is real, here's the proof."

Quick start

With uv installed, no global Python package install is required. Install Chromium once:

uvx --from playwright playwright install chromium

Then connect Argus to your MCP client.

Claude Code

claude mcp add argus -- uvx --from argus-testing argus-mcp

Codex and the ChatGPT desktop app

Codex CLI, the Codex IDE extension, and the ChatGPT desktop app share the same local MCP configuration:

codex mcp add argus -- uvx --from argus-testing argus-mcp

Cursor

The button adds Argus to Cursor; run the Chromium installation command above once before the first test.

Any stdio MCP client

{
  "mcpServers": {
    "argus": {
      "command": "uvx",
      "args": ["--from", "argus-testing", "argus-mcp"]
    }
  }
}

The default core profile exposes the primary web-testing workflow without flooding the host with every specialist tool. Use uvx --from argus-testing argus-mcp --list-tools to inspect the selected profile, --tool-profile screen for native macOS testing, or --tool-profile full for the entire advanced surface. ARGUS_TOOL_PROFILE provides the same setting through the environment.

Then just ask, in your agent session:

"Test my app at http://localhost:3000 — find real bugs."

That's it. The agent drives; Argus keeps it honest and writes the report.

For a scoped review, the host can give start_session explicit goals, constraints, and an advisory time_budget_minutes. Argus returns the full testing protocol once and keeps outstanding goals and discovered pages visible in later observations. Mark a goal in_progress before its journey; when coverage_update marks it exercised or blocked, Argus requires a concrete explanation and automatically links the URLs, value-redacted actions, screenshots, persistence checks, bugs, and observations produced in that testing window. The final HTML and JSON reports preserve both completed and unfinished coverage instead of implying that an incomplete pass was comprehensive.

pip install argus-testing
playwright install chromium
claude mcp add argus -- argus-mcp

# Uses a LiteLLM-backed planner. Set a provider key (OPENAI_API_KEY, DEEPSEEK_API_KEY, …).
uvx --from argus-testing argus http://localhost:3000 --model deepseek/deepseek-chat

# Higher recall: union N independent passes (deduped, proven instance kept)
uvx --from argus-testing argus http://localhost:3000 --passes 3

pip install 'argus-testing[mac]'
brew install cliclick          # keystroke / coordinate fallback
argus-mcp --doctor             # check Screen Recording + Accessibility grants
claude mcp add argus-screen -- argus-mcp --tool-profile screen

Same description-keyed tools, but the target is whatever app is foreground on macOS — Notes, Cursor, Safari, your in-progress feature. No headless Chrome, no scripted Playwright. Argus sees what you see, via the Accessibility tree.

Why Argus is different

Existing testing tools only test what you script. Playwright and Cypress run the assertions you wrote. Argus discovers bugs you didn't think to test for — and then does the thing an LLM alone can't be trusted to do: proves them.

How it compares

On the axis that matters for finding bugs — autonomously discover, independently verify, and report — Argus occupies a different slot from the browser-MCP crowd:

These aren't "worse" tools — they're a different job. Playwright MCP gives an agent excellent hands; Chrome DevTools MCP gives it deep network/perf/memory inspection Argus doesn't have. Argus is the layer that decides what's a bug and proves it. Use them together.

Benchmarks

$ python -m argus.bench --target all

  buggytasks    22 / 22  = 100 %   ·  mechanical bugs (console errors, fake delete, auth bypass…)
  darkshop      12 / 12  = 100 %   ·  human-eye bugs (fake scarcity, lying toasts, stale state…)
  ──────────────────────────────────────────────────────────────────────
  total         34 / 34  = 100 %   ·  reproducible from git clone in two commands

34 / 34 is the capability ceiling — what's findable through the tool surface, measured by deterministic scripts. It is deliberately separate from how often a given LLM remembers to use the tools well, which is noisy and honestly reported below.

python -m argus.bench.agent_runner puts an actual model in the driver's seat and scores recall across trials. What we've learned running it:

1. Real recall sits well below the 34/34 ceiling. A live driver finds a fraction of the seeded bugs per pass — the ceiling is what's findable, this is what a model finds.

2. Variance is large — never rank models on a few runs. Per-trial recall swings widely; we report the spread, not a single hero number.

3. Dogfooding the bench found real bugs in Argus itself — a record_bug crash on a string argument that silently dropped findings, resolver misses on common phrasings. The tool-testing tool got tested.

4. Precision holds regardless of driver. Across every trial, the reproduction receipt kept false-certifications at zero — a weak model finds fewer bugs, but the ones marked VERIFIED are still real.

BuggyTasks (:5555) — 22 mechanical bugs in a task app: console errors, dead links, fake delete (UI says "deleted!" but data persists on refresh), auth bypass, NaN dates, off-by-one counts, race conditions. The "scripted E2E could find these" tier.

DarkShop (:5556) — 12 human-eye bugs in a polished-looking store: hardcoded "Only 3 left!" scarcity, -50% badges where sale price equals original, a "free shipping over $50" banner contradicted by a flat $5 at checkout, inverted visual hierarchy ("Add to Cart" demoted under a prominent "Subscribe"), cross-page state drift (rename sticks on /account, navbar greeting doesn't). Static analysis catches roughly none of these — they require an agent that reads the page and reasons.

python test-site/app.py           # BuggyTasks  :5555
python human-eye-fixture/app.py   # DarkShop    :5556
python -m argus.bench --target all

Tool surface

argus-mcp starts with the focused core web profile. Every public tool is documented below. The counts are also available directly from the installed server:

uvx --from argus-testing argus-mcp --list-tools
uvx --from argus-testing argus-mcp --tool-profile screen --list-tools
uvx --from argus-testing argus-mcp --tool-profile full --list-tools

Reports keep original screenshots as evidence and, by default, write compact WebP previews under report-assets/ instead of base64-embedding every full-size PNG into the HTML. Set ARGUS_PORTABLE_REPORT=1 when a single self-contained HTML file is more important than size. JSON output includes complete reproduction receipts, the coverage contract and its structured evidence references, constraints, review mode, tool-call and recorded-step counts, screenshot metadata, and qualitative observations. JUnit suite failure totals match the emitted <failure> nodes.

Safety: per-call timeout, a 30-minute session cap, a ~/.argus/abort panic file that halts every subsequent action, and an automatic before/after screenshot trail on every action.

The full profile includes every core and screen tool above plus these 36 specialist tools. Use it when the workflow genuinely needs low-level state, fault injection, multi-tab control, coordinates, or crawling.

To expose eval_js as an operational tool rather than a disabled safety stub:

uvx --from argus-testing argus-mcp --tool-profile full --unsafe

Local-first security and privacy

Argus runs on your machine and does not send telemetry to an Argus-operated service. Reports and screenshots stay under ./argus-reports by default; your MCP host and its configured model provider can still receive tool results included in the conversation. Browser actions and native macOS controls can cause real side effects, so use test accounts and non-production data wherever possible.

Read the full privacy disclosure and security policy before using Argus against sensitive systems.

Philosophy

Argus assumes an Opus-class driver. Static rules that pretend to be the smart layer are subtractive — they add maintenance and false positives and pull attention from what the agent actually saw. So detector.py is tiny: it only captures the two channels the agent literally cannot see (the console event stream and the HTTP layer). "Is this toast misleading? Is the visual hierarchy wrong? Is that count off?" — the agent reads observe() and decides.

The global instruction is intentionally tiny so it does not repeat a long QA prompt in every MCP tool description. start_session returns the full evidence-first ritual, goals, constraints, and budget once; observations then surface only the compact live coverage ledger. Argus remains a capability inside the user's current task: it does not prevent implementation work, replace the host's identity, or imply authority for irreversible external actions.

click_what("Login button"), not click(7). Element indices are a leaky abstraction even within one observe. A capable agent describes what it wants by what it is, and the resolver maps that to the right element — refusing to misclick on ambiguity rather than guessing.

Project layout

argus/
├── mcp_server.py     # tool surface + role instructions + reproduction-receipt engine
├── browser.py        # Playwright backend: DOM/ARIA extraction, capsule/replay
├── resolver.py       # description → element (web + screen)
├── reporter.py       # HTML + JSON + JUnit + SARIF
├── detector.py       # console + network capture (only)
├── cli.py            # argus (explore) + argus-regression
├── bench/            # deterministic ceiling + real-LLM recall harness
└── screen/           # macOS AX backend, permissions, safety
test-site/            # BuggyTasks  (22 mechanical bugs)
human-eye-fixture/    # DarkShop    (12 human-eye bugs)

MIT licensed · Product page · Agent install guide · Privacy · Security · Built by Yichen Wu