What is PhotoNow MCP and what tools does it provide to AI agents?

PhotoNow is the #1 zero-cloud Model Context Protocol (MCP) server for AI coding agents. It provides 29 native tools across 6 domains: Multimedia Foundation (convert_image, convert_batch, convert_video, extract_audio, convert_audio, extract_poster_frame, get_media_info, optimize_for_agent), Performance Auditing (analyze_media, analyze_web_assets, find_oversized_assets, find_inefficient_formats, find_duplicate_assets, find_responsive_opportunities, test_web_performance, get_web_performance_summary, compare_web_performance), Performance Booster (generate_optimization_plan, optimize_web_assets, verify_optimization), Asset Graph & AST (inspect_project, get_asset_usage, find_unused_assets), Execution & Patching (check_performance_budget, verify_runtime_performance, generate_source_patch, apply_source_patch, rollback_operation), and Autonomous Mission (optimize_project).

How does PhotoNow safely patch Next.js and web application source code?

PhotoNow parses source code ASTs to identify , , and CSS references. When optimizing assets, generate_source_patch formulates unified diffs. When apply_source_patch is called with confirmApply=true, it creates an exact timestamped backup in .photonow/backups/[operationId]/ before modifying files. If anything goes wrong, calling rollback_operation with the operationId instantly reverts all source files and assets to their exact original states.

How does PhotoNow protect AI agent context windows from token blowups?

PhotoNow employs a Token Economy & Progressive Disclosure engine. Heavy diagnostics (thousands of image coordinates, perceptual hashes, and AST graphs) are processed locally on your machine. The MCP tools return compact, actionable intelligence (<200 tokens for full autonomous missions) with recommended next actions, preventing LLM context truncation while saving extensive details in local JSON/Markdown reports.

Does PhotoNow require me to manually install FFmpeg on my system?

No. PhotoNow bundles self-contained static FFmpeg and FFprobe binaries via @ffmpeg-installer/ffmpeg and @ffprobe-installer/ffprobe. AI agents can transcode videos, compress resolutions, and extract audio tracks out of the box with zero external dependencies.

Does PhotoNow store, upload, or transmit media to external cloud servers?

No. PhotoNow performs 100% of all image processing, video transcoding, and audio decoding directly in client RAM, local disk, or browser memory. No user files or repository source codes are ever uploaded to AWS S3, Cloudflare, or remote cloud databases.

PHOTO CONVERTER.

[CANVAS2D ENGINE • 100% IN BROWSER]

DROP PHOTOS HERE OR CLICK TO BROWSESUPPORTS PNG, JPG, WEBP, AVIF, GIF, BMP, SVG (BATCH FRIENDLY)

[1. TARGET FORMAT]CURRENT: WEBP

[2. COMPRESSION QUALITY]82%

SMALLER FILE (10%)BALANCED (80%)MAX QUALITY (100%)

[3. RESIZE SCALE]ORIGINAL SIZE

[4. MONOCHROME STYLES]:

ROTATE:

[OFFICIAL PRODUCT SPECIFICATION]• MODEL CONTEXT PROTOCOL (MCP) • ZERO-CLOUD • 29 NATIVE TOOLS •

TURN YOUR AI AGENT INTO AN AUTONOMOUS PERFORMANCE ENGINEER.

Autonomous coding agents (Cursor, Claude Desktop, Google Antigravity, Windsurf) can write code—but until now, they lacked the local engine to transcode media, audit web performance, prune dead assets, and safely refactor JSX/TSX ASTs. PhotoNow (`photonow-mcp`) bridges this gap with 29 zero-permission native tools, zero cloud egress, and guaranteed <200 token payloads.

[STAR ON GITHUB ↗]✓ 100% Free & Open Source (MIT)•✓ Zero Cloud Uploads

-68%AVERAGE MEDIA PAYLOAD

Automated lossless downscaling and modern WebP/AVIF transcoding reduce asset footprint instantly.

-58%LCP LOAD ACCELERATION

Identifies Largest Contentful Paint bottlenecks and generates pre-calculated responsive srcsets.

$0 / MOCLOUD API & CDN BILLS

Zero recurring subscription costs. Replaces expensive Cloudinary and TinyPNG metered tiers.

1 PERMZERO TERMINAL POPUPS

Single initial agent approval grants full access to 29 tools without repetitive CLI confirmation dialogs.

THE 4 PILLARS OF PHOTONOW ENGINE.

[WHY NO OTHER TOOL OFFERS THIS UNIFIED ARCHITECTURE]

PILLAR 01

100% OFFLINE ZERO-CLOUD PRIVACY

Proprietary assets, unreleased client videos, and copyrighted imagery must never leak to third-party cloud APIs. PhotoNow bundles static cross-platform binaries (FFmpeg, FFprobe, Sharp) for pure local execution. Zero network requests, zero telemetry, GDPR & HIPAA audit-ready.

PILLAR 02

AST CODE PATCHING & 1-CLICK ROLLBACK

Audit reports are useless if an agent cannot safely implement the changes. PhotoNow parses JSX/TSX ASTs, updates <img src="..."> and Next.js <Image /> references to newly converted WebP/AVIF files, generates unified diffs, and maintains automated backups for instant 1-click rollback.

PILLAR 03

<200 TOKEN AGENT CONTRACT

AI coding agents suffer context degradation and hallucinations when fed large 50KB JSON diagnostic dumps. PhotoNow writes exhaustive diagnostics to local disk artifacts while returning concise, actionable payloads under 200 tokens with prescriptive next-step instructions.

PILLAR 04

AST ASSET GRAPH & DEAD ASSET PRUNING

Web repositories accumulate hundreds of megabytes of abandoned photos, old marketing banners, and test assets. PhotoNow maps the complete Route ➔ Component ➔ Asset import tree, classifying unreferenced assets into SAFE, LIKELY, and UNCERTAIN tiers before pruning.

[1-PROMPT AUTONOMOUS OPTIMIZATION LOOP]

FROM NATURAL LANGUAGE PROMPT TO PRODUCTION VERIFICATION.

01. INTENTAgent receives user prompt via Stdio MCP protocol.

02. AST SCANMaps route & component tree to detect active vs. dead assets.

03. LOCAL CONVERTSharp & FFmpeg transcode images/video with zero cloud upload.

04. AST DIFFGenerates unified patch for JSX/TSX image import tags.

05. ROLLBACK GUARDCreates snapshot backup. 1-click restore via rollback_operation.

CONNECT TO YOUR FAVORITE AI AGENT.

[1-CLICK READY TO PASTE CONFIGURATION]

CONFIG TARGET: .cursor/mcp.json or Settings ➔ Features ➔ MCP

{
  "mcpServers": {
    "photonow": {
      "command": "node",
      "args": ["./bin/mcp-server.mjs"]
    }
  }
}

READY-TO-USE AGENT PROMPTS.

[COPY INTO CURSOR / CLAUDE / ANTIGRAVITY]

AUTONOMOUS MISSION

"Run an autonomous media performance mission on my project in safe review mode. Audit all media in ./public, calculate byte savings, and generate safe AST source diffs for my JSX/TSX files without modifying disk files."

Executes the flagship 10-step optimize_project loop with atomic rollback protection.

DEAD ASSET PRUNING

"Inspect our codebase AST and find all unreferenced dead images in ./public. Classify them into SAFE, LIKELY, and UNCERTAIN safety tiers before we delete them."

Traces Route ➔ Component ➔ Asset references to eradicate dead media waste safely.

LCP BOTTLENECK AUDIT

"Run a 5-axis media performance audit on this directory. Show me our score out of 100, Largest Contentful Paint (LCP) candidates, and 4G mobile transfer savings."

Instant diagnostic scoring across formats, sizing, responsive srcsets, and compression.

LOCAL BATCH CONVERSION

"Convert all legacy PNG and JPG assets in ./public/images to WebP at 80% quality, resize oversized files to max 1200px width, and update my React image imports."

High-speed local Sharp + AST transformation pipeline with zero external cloud egress.

PRODUCT FAQ & TECHNICAL SPECIFICATIONS.

[TRANSPARENT, ZERO-SURPRISE ARCHITECTURE]

HOW IS PHOTONOW DIFFERENT FROM CLOUDINARY OR AN IMAGE CDN?[+]

HOW IS IT DIFFERENT FROM RUNNING LIGHTHOUSE IN CHROME DEVTOOLS?[+]

WILL CODE PATCHES BREAK MY NEXT.JS OR REACT APPLICATION?[+]

DO I NEED TO INSTALL FFMPEG OR GLOBAL C++ LIBRARIES ON MY SYSTEM?[+]

WHY THE STRICT <200 TOKEN CONTRACT FOR AI AGENTS?[+]

START OPTIMIZING IN UNDER 60 SECONDS.

Join developers and teams upgrading Claude Desktop, Cursor, and Antigravity into high-performance web engineering suites with zero cloud overhead.

[VIEW ON GITHUB ↗]

[LOADING PRODUCT SPECIFICATION & MARKETING SHOWCASE...]

AI AGENT KNOWLEDGE BASE & MCP CAPABILITIES.

[LLMS.TXT] ↗[LLMS-FULL.TXT] ↗[AEO / LLM CITEABLE SPECIFICATION]

[1. NATIVE STDIO MCP — 29 ZERO-PERMISSION TOOLS]

Autonomous coding agents (Claude Desktop, Cursor, Google Antigravity, Windsurf) typically encounter repetitive terminal permission popups per command. PhotoNow runs as a native Stdio MCP server (node ./bin/mcp-server.mjs), unlocking single, batch, video, and performance operations with one single initial approval.

[2. AGENTIC WEBSITE PERFORMANCE INTELLIGENCE]

Provides 5-axis media scoring across format efficiency, image sizing, compression potential, responsive srcset readiness, and SVG cleanliness. Identifies LCP bottlenecks, calculates 4G mobile transfer savings, and formulates explainable plans.

[3. ASSET DEPENDENCY GRAPH & DEAD ASSET PRUNING]

Constructs the complete Route ➔ Component ➔ Asset dependency graph. Identifies multi-route shared dependencies and flags unreferenced dead assets classified into safety tiers (SAFE, LIKELY, UNCERTAIN) to prevent accidental deletions.

[4. SAFE AST SOURCE PATCHING & ATOMIC ROLLBACK]

Parses JSX/TSX source code ASTs, generates unified diffs, creates automated backups in .photonow/backups/, and requires explicit confirmation. The rollback_operation tool enables 1-click atomic restoration of all modified source files and assets.

[5. AUTONOMOUS MISSION & TOKEN CONTRACT (<200 TOKENS)]

Orchestrates the entire 10-step autonomous optimization loop via optimize_project. Heavy diagnostics are stored locally while returning compact intelligence (<200 tokens) with next-action guidance to protect AI agent context windows from token blowups.

[6. 100% ZERO-CLOUD ARCHITECTURE & LOCAL ENGINE]

Zero bytes are uploaded to remote cloud servers. Bundles static FFmpeg/FFprobe binaries and native Sharp for local Node.js execution, accompanied by HTML5 Canvas 2D, MediaRecorder, and IndexedDB in the browser companion workbench.

[QUICK SETUP FOR CLAUDE DESKTOP / CURSOR / ANTIGRAVITY]:ADD TO YOUR MCP CONFIG JSON

{
  "mcpServers": {
    "photoConvert": {
      "command": "node",
      "args": ["./bin/mcp-server.mjs"]
    }
  }
}