RenderProof
A local-first MCP server that captures rendered web evidence including screenshots, motion, pixel diffs, and CSS animation metadata for coding agents.
README
RenderProof
Rendered web evidence for coding agents.
RenderProof is a local-first MCP server that lets Codex, Claude Code, Cursor, and other agents capture what Chromium actually rendered: screenshots, motion, CSS animation metadata, pixel diffs, gates, loading states, and visual evidence that plain fetch tools miss.
It is not a scraper. It is not a bypass tool. It is the receipt for what the browser saw.
Why This Exists
Coding agents are very good at reading source. The web is increasingly bad at being understood from source alone.
Modern pages hide their real state behind client rendering, consent walls, cookie modals, skeleton loaders, canvas, video surfaces, iframes, maps, bot checks, and motion. A text fetch can say "Google Maps"; a browser screenshot can show "Danish consent wall blocking the map."
RenderProof gives agents a grounded visual evidence layer before they make claims or copy a design.
What It Does
| Tool | Use it when | Output |
|---|---|---|
route_web_task |
The agent needs to choose text, screenshot, motion, search, or another platform tool | Route recommendation with evidence |
read_url |
Text is enough | Normalized direct fetch text, with optional remote reader |
capture_page |
Layout, gates, rendered state, canvas, images, maps, PDFs-as-browser-state, or visual QA matter | PNG screenshot + metadata |
capture_motion |
A human needs to see animation, loading, scroll, transitions, or media motion | WebM video + keyframe PNGs |
analyze_motion |
A coding agent needs to recreate or critique motion | Contact sheet, diff image, CSS animation metadata, pixel-diff JSON, design notes |
doctor |
You want to verify local runtime/browser setup | Runtime and Playwright checks |
The Core Idea
flowchart LR
A["Agent task"] --> B["route_web_task"]
B --> C["read_url"]
B --> D["capture_page"]
B --> E["capture_motion"]
B --> F["analyze_motion"]
C --> G["Grounded answer"]
D --> G
E --> G
F --> G
RenderProof makes the expensive path explicit: read text first when text is enough, escalate to rendered evidence when pixels matter.
Example Prompts
Use RenderProof to capture a full-page screenshot of https://example.com with autoScrollBeforeCapture true.
Use RenderProof to analyze the animation style on this page and return the contact sheet, pixel-diff summary, and CSS animation metadata.
Use RenderProof to check what the browser actually sees on this Instagram/TikTok/YouTube page.
Use RenderProof to capture visual evidence before claiming whether this page is blocked by a consent wall, login wall, CAPTCHA, or skeleton loader.
Motion Analysis
Video is useful for humans. Agents need structured evidence.
analyze_motion samples frames over time and returns:
- a contact sheet PNG
- a red-highlight diff PNG
- frame timestamps
- changed-pixel ratios
- changed-region bounds and centroid movement
- dominant movement direction
document.getAnimations()metadata- CSS keyframes, duration, easing, direction, iterations, target selector, animated properties
- design notes written for a coding agent
Example design notes:
[
"Detected 2 CSS/Web Animation target(s), affecting transform.",
"Primary timing appears to use duration 1200ms, easing ease-in-out, direction alternate.",
"Pixel sampling saw up to 5.81% of sampled pixels change between adjacent frames.",
"The changed region centroid trends right across sampled frames."
]
That is the difference between "there is an animation" and "recreate this with transform, 1200ms, ease-in-out, alternate, moving right while scaling."
Install
One-liner:
curl -fsSL https://raw.githubusercontent.com/Djsand/renderproof/main/install.sh | bash -s -- codex
Windows PowerShell:
powershell -NoProfile -ExecutionPolicy Bypass -Command "& ([scriptblock]::Create((irm https://raw.githubusercontent.com/Djsand/renderproof/main/install.ps1))) -Target codex"
Swap codex for claude, cursor, cline, windsurf, gemini, or generic.
The codex target writes directly to ~/.codex/config.toml, so it works even when the Codex CLI is not available in your terminal PATH.
If Windows says Missing required command: node, install Node.js 20+ and reopen PowerShell:
winget install OpenJS.NodeJS.LTS
Manual install:
git clone https://github.com/Djsand/renderproof.git
cd renderproof
npm install
npx playwright install chromium
Then install it into your coding agent:
node dist/index.js install
One-liners:
curl -fsSL https://raw.githubusercontent.com/Djsand/renderproof/main/install.sh | bash -s -- codex
curl -fsSL https://raw.githubusercontent.com/Djsand/renderproof/main/install.sh | bash -s -- claude
curl -fsSL https://raw.githubusercontent.com/Djsand/renderproof/main/install.sh | bash -s -- cursor
curl -fsSL https://raw.githubusercontent.com/Djsand/renderproof/main/install.sh | bash -s -- cline
curl -fsSL https://raw.githubusercontent.com/Djsand/renderproof/main/install.sh | bash -s -- windsurf
curl -fsSL https://raw.githubusercontent.com/Djsand/renderproof/main/install.sh | bash -s -- gemini
Windows PowerShell one-liners:
powershell -NoProfile -ExecutionPolicy Bypass -Command "& ([scriptblock]::Create((irm https://raw.githubusercontent.com/Djsand/renderproof/main/install.ps1))) -Target codex"
powershell -NoProfile -ExecutionPolicy Bypass -Command "& ([scriptblock]::Create((irm https://raw.githubusercontent.com/Djsand/renderproof/main/install.ps1))) -Target claude"
powershell -NoProfile -ExecutionPolicy Bypass -Command "& ([scriptblock]::Create((irm https://raw.githubusercontent.com/Djsand/renderproof/main/install.ps1))) -Target cursor"
powershell -NoProfile -ExecutionPolicy Bypass -Command "& ([scriptblock]::Create((irm https://raw.githubusercontent.com/Djsand/renderproof/main/install.ps1))) -Target cline"
powershell -NoProfile -ExecutionPolicy Bypass -Command "& ([scriptblock]::Create((irm https://raw.githubusercontent.com/Djsand/renderproof/main/install.ps1))) -Target windsurf"
powershell -NoProfile -ExecutionPolicy Bypass -Command "& ([scriptblock]::Create((irm https://raw.githubusercontent.com/Djsand/renderproof/main/install.ps1))) -Target gemini"
Local helper equivalents:
node dist/index.js install codex --write-user
node dist/index.js install claude --apply
node dist/index.js install cursor --write-project
node dist/index.js install cline --write-user
node dist/index.js install windsurf --write-user
node dist/index.js install gemini --apply --scope user
See docs/installation.md for agent-specific setup.
Use With Codex
node dist/index.js install codex --write-user
Manual equivalent:
codex mcp add renderproof -- node /absolute/path/to/renderproof/dist/index.js mcp
Direct config equivalent:
[mcp_servers.renderproof]
command = "node"
args = ["/absolute/path/to/renderproof/dist/index.js", "mcp"]
Use With Claude Code
node dist/index.js install claude --apply
Manual equivalent:
claude mcp add renderproof -- node /absolute/path/to/renderproof/dist/index.js mcp
Use As CLI
npm run dev -- route "summarize this page" --url https://example.com
npm run dev -- read https://example.com
npm run dev -- capture https://example.com --full-page
npm run dev -- capture https://example.com --full-page --auto-scroll
npm run dev -- motion https://example.com --duration 5000 --keyframes true
npm run dev -- analyze-motion https://example.com --duration 3000 --samples 5
npm run dev -- install generic --json
npm run doctor
Safety Model
RenderProof is designed to show gates, not bypass them.
By default, it:
- allows public
httpandhttpsURLs - blocks
localhost, private IP ranges, and hosts resolving to private IPs - does not use remote readers unless explicitly enabled
- does not solve CAPTCHA
- does not hide automation
- does not click through consent, login, or payment walls
- returns structured failures instead of pretending a page was accessible
Environment controls:
RENDERPROOF_ALLOWED_HOSTS=example.com,*.example.org
RENDERPROOF_ALLOW_PRIVATE_NETWORK=1
RENDERPROOF_ENABLE_REMOTE_READERS=1
RENDERPROOF_OUTPUT_DIR=.renderproof/evidence
RENDERPROOF_MAX_CHARS=12000
JINA_API_KEY=...
Legacy GROUNDED_WEB_* variables are still supported for compatibility.
What RenderProof Is Good At
- Seeing rendered browser state when direct fetch returns useless HTML
- Capturing cookie modals, login walls, consent walls, reCAPTCHA widgets, and skeleton loaders
- Capturing maps, video players, image pages, canvas/WebGL surfaces, and JS-computed UI
- Producing full-page screenshots with optional pre-scroll for lazy loading
- Recording short WebM motion evidence for humans
- Turning animation into agent-readable CSS and pixel-diff evidence
- Creating auditable local evidence files for coding workflows
What It Will Not Do
- bypass paywalls, login walls, CAPTCHA, bot checks, or private networks by default
- scrape social platforms as a data product
- replace Playwright MCP for interactive browser control
- guarantee semantic object tracking from pixel diffs alone
Playwright is the engine. RenderProof is the evidence layer.
Development
npm install
npm run typecheck
npm run build
npm run doctor
Run a browser launch check:
npm run dev -- doctor --check-browser-launch
Roadmap
- PDF rendering and text extraction
- page-state detection for consent/login/CAPTCHA/skeleton/download/blank-page states
- style-reference capture for "build me a site like this" workflows
- safe user-provided auth/session support
- evidence bundle manifests
- optional interaction sequences with before/after evidence
License
MIT
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.