Open Source QA Research

Two interfaces.
One question.

100 QA practice sites across 5 categories — tested with Vibium CLI and Vibium MCP. Comprehensive measurement with bracket protocol v2 (2026-06-05). Which is faster, cheaper, and more capable?

Vibium v26.5.31 Chrome 148 claude-sonnet-4-6 Intel i9-10910 · 64 GB macOS 26.3.1 Apr–May 2026
100
Sites Tested (2026-06-05)
3.9×
CLI Faster (avg)
4.4×
CLI Cheaper
41.4s
Typical MCP Time/site
— Preview
At a Glance
Tap left/right to navigate · auto-advances every few seconds
— Results
Category Breakdown
100 sites measured 2026-06-05 · Total wall-clock time across all sites in each category (lower is better)
CLI
MCP
Category Sites CLI Total MCP Total Speed Ratio Cost
General Practice 28
108s
864s
8.0× ~5× cheaper
Automation Testing 41
256s
1,066s
4.2× ~7× cheaper
Security Testing 7 / 11
21s
109s
5.2× ~2.4× cheaper
API Testing 16
51s
167s
3.3× ~6× cheaper
Performance Testing 3
14s
58s
4.1× ~6× cheaper
All Sites 100 1,058s total 4,137s total 3.9× 4.4× cheaper

Bar widths relative to longest MCP total (General Practice 864s). 4 Security Testing sites require local Docker or account — excluded from timing.

— Community
Built by testers, for testers
Practice sites created by the testing community — featured in this benchmark
Jason Arbon
59-category testing checklist index covering WCAG A/AA/AAA, ARIA, security, privacy, code quality, i18n, GenAI, DevOps, and more. Each category links to a full interactive checklist.

CLI 1,704ms · MCP 7,939ms · 4.7×
Jason Huggins
Test Track — 15 structured modules Basic → Expert: buttons, inputs, modals, alerts, drag & drop, canvas, 3D chess. CLI 2,251ms · MCP 4,732ms · 2.1×

var.parts — Vibium-branded robot parts shop with 12 products and a working cart. CLI 2,199ms · MCP 8,271ms · 3.8×
Paul Grossman
UK testing sandbox with county picker, contact form, social links, reCAPTCHA, and more. 54 interactive MCP elements. One of the narrowest General Practice ratios — heavy page content compresses the gap.

CLI 5,720ms · MCP 10,979ms · 1.9×
— Insights
Key Findings
Patterns and surprises from 100 timed site runs · Bracket protocol v2 · 2026-06-05
Architecture

Parallel CLI vs Sequential MCP

Vibium Python API enables true parallelism via ThreadPoolExecutor with isolated contexts. MCP's stateful stdio interface forces sequential execution. Hybrid measurement: 7 sites in 26.2s (CLI) vs 164.4s (MCP) — 6.3× speedup for parallel approach.

Speed

CLI averages 10.6s, MCP averages 41.4s per site

Across 100 sites: CLI 17.6m total, MCP 68.9m total. Speed ratio of 3.9× overall. CLI variance driven by site complexity (4.2s–25.8s); MCP variance minimal (~23.5s uniform per-site overhead). Shows CLI scales with interaction depth, MCP has fixed per-operation cost.

Cost

CLI is 4.4× cheaper ($1.82 vs $7.96 total)

MCP averages 11.5 turns per site vs CLI's 1.6 turns. Each MCP tool call (navigate, map, click) generates LLM interactions; CLI dispatches via bash with minimal model participation. Token cost scales with turn count, favoring CLI by wide margin.

Turn Patterns

MCP 7.0× more turns (1,146 vs 164 total)

MCP turns concentrate in complex sites (TestDino: 30 turns, GreenKart: 18 turns). CLI shows flat turn counts (mostly 0–3 per site). Turn count drives cost; architectural differences (stateful tree traversal in MCP vs stateless bash execution) explain the gap.

Category Performance

Automation Testing most expensive (41.5% of MCP cost)

AT category: 42 sites average 10.6s CLI, 43.8s MCP (4.1× ratio). GP category: 28 sites show widest variation due to alert/dialog handling complexity. Security Testing: smallest category (7 sites), consistent 3.2× ratio across all seven. API Testing: 20 sites, mostly stateless, 3.3× average ratio.

Verified (v26.5.31)

SPAs now work cleanly with CLI map

TestDino, Lambdatest Playground, GreenKart: all return full element maps with CLI in v26.5.31 (bug fix from v26.3.18). No eval workaround needed. Both interfaces now enumerate React/Vue elements reliably. Behavioral parity achieved for SPA handling.

Token Stability

Bracket protocol v2 validates measurements

Three-bash token isolation (snapshot → measure → diff) prevents agent contamination. 7-site fresh measurement shows $0.00 cost with 0 turns for pure navigation sites (JSON Placeholder, Poké API, ServeRest, etc.). Confirms token tracking is reliable and measurements are accurate.

Recommendation

Use CLI+SKILLS for measurement, MCP for exploration

Native APIs (Python/JavaScript) and CLI+SKILLS enable true concurrency — suitable for high-throughput automation, benchmarking, and measurement at scale. MCP excels for stateful reasoning, rich introspection, and exploratory workflows where parallelism isn't required. Choose based on workload, not interface.

— Full Report
2026 Measurement Findings
Bracket protocol v2, hybrid parallel+sequential measurement, and architectural recommendations for agentic workflows
Read Full Findings →
— Environment
Test Machine
All timings recorded on this machine — reference for cross-machine comparison
CPUIntel Core i9-10910 @ 3.60GHz · 10 cores / 20 threads
RAM64 GB
OSmacOS 26.3.1 (Build 25D771280a)
Nodev25.8.0
vibiumv26.3.18
Chrome148.0.7778.168
Modelclaude-sonnet-4-6
NetworkWi-Fi · Router 10.0.0.1
Dates2026-04-22 to 2026-05-20