Skip to content
ComposeProof

Introduction

ComposeProof

Section titled “ComposeProof”

ComposeProof is an MCP server that gives AI coding assistants eyes on Compose UI. It provides headless rendering, embedded runtime inspection, golden management, spec-driven verification, device interaction, and expert analysis prompts.

Works with Claude Code, Gemini CLI, Cursor, Android Studio Agent Mode — anything that speaks MCP.

What it does

Section titled “What it does”
You: "verify my login screen against this spec"
AI:  reads spec, maps assertions to @Preview functions
AI:  calls verify → render + golden check + accessibility
AI:  "PASS (3/3) — rendered 1.8s, 99.8% golden match, 0 a11y warnings"

At scale:

AI:  calls render_batch mode='verify'
AI:  "50 previews: 45 PASS, 3 FAIL, 2 NEW"
AI:  drills into failures, suggests fixes, re-verifies

No device. No emulator. The AI IS the test engine — ComposeProof gives it eyes.

Key capabilities

Section titled “Key capabilities”
  • 40 MCP tools across headless rendering, device interaction, runtime inspection, code intelligence, testing, and mocking
  • 11 expert prompts (8 free + 3 Pro) for accessibility, performance, KMP architecture, UI review, Compose best practices, Rebound budgets, stability diagnosis, and migration
  • Zero-install — works on any Compose project without build file changes
  • Headless rendering via Compose Desktop (Skia) — no device, no emulator needed
  • Golden management — record, verify, and update visual baselines
  • Embedded agent — runtime inspection of Compose state, navigation, coroutines, ViewModel, and more
  • Mock API server — intercept real API calls with mock responses, no code changes
  • Edge-case generation — analyze composable parameters and suggest visual test cases
  • HTML reports — self-contained reports with screenshots, spec assertions, and context graph

Who it’s for

Section titled “Who it’s for”
  • Android/Compose developers using AI coding assistants (Claude Code, Gemini CLI, Cursor) who want the AI to see and verify UI output
  • Teams wanting automated visual regression testing in CI/CD pipelines
  • Solo developers who need QA automation without hiring QA

How it works (30 seconds)

Section titled “How it works (30 seconds)”
  1. Install: npx composeproof
  2. Open your AI assistant in any Compose project
  3. Ask: “render my HomeScreen preview”
  4. ComposeProof renders it headlessly via Skia, the AI sees the screenshot
  5. Ask: “does it match the spec?” — the AI verifies and reports back

No device. No emulator. No build file changes. The AI closes the feedback loop.

What’s inside

Section titled “What’s inside”
CategoryCountExamples
Context & Orientation4insights, get_context, configure_context, generate_report
Headless Rendering6render, list_previews, verify, render_batch, diff
Device Interaction5preflight, build_and_deploy, device_interact, take_screenshot
Device Inspection5inspect_ui_tree, get_network_logs, get_recomposition_stats
Embedded Agent — Runtime7inspect_permissions, inspect_navigation_graph, simulate_process_death
Embedded Agent — Compose5semantic_ui_query, inspect_compose_state, track_recompositions
Embedded Agent — App Internals4inspect_shared_preferences, inspect_viewmodel_state, inspect_network_logs
Code Intelligence (Pro)2index_project, semantic_search
Testing & Mocking2generate_edge_cases, mock_api
Expert Prompts (Free)8accessibility-checker, compose-skill, rebound-basics
Expert Prompts (Pro)3stability-doctor, rebound-advisor, migration-assistant

Next steps

Section titled “Next steps”