SoloMCN – Claude Code as the runtime for a one-person video studio

Developer ymybxx published SoloMCN, an open-source MIT-licensed video production workbench that turns Anthropic's Claude Code CLI into an autonomous one-person media channel network, according to github.com's report. Instead of wiring rigid multi-agent graphical workflows with hardcoded prompts, the software treats Claude Code as a seated employee and provides a local workstation interface. The runtime connects hot-topic scraping across platforms like Douyin, Weibo, Bilibili, Zhihu, Baidu, Toutiao, Hacker News, X, and YouTube directly to autonomous scriptwriting, dynamic asset capture, video assembly, and cross-platform publishing.

This technical breakdown covers the underlying architecture, execution pipeline, local skill management, and system setup of SoloMCN. The software is designed for independent technical creators, developer advocates, and solo video operators who want to produce research-backed technical explainers or stage montages without manually editing timelines. Operating locally on macOS, the system requires only a machine environment, a Claude subscription, and a standard browser session to automate short-form video creation while leaving editorial approval gates entirely in human hands.

What shipped in SoloMCN – Claude Code as the runtime

The release delivers a complete local workbench comprising a vanilla HTML, CSS, and JavaScript interface in public/, an Express-style backend in server/, a dedicated Python hot-topic aggregator in hot-service/, and browser automation scripts in publish/. The system replaces conventional node-based workflow engines with direct agent invocation through server/agent.js, which spawns the Claude Code CLI in headless mode using claude -p. An internal Model Context Protocol module in server/mcp.js exposes local file read/write, topic recording, script storage, and web-page screenshot capabilities directly to Claude. Platform scraping includes bilisource.js for fetching 1080p clips and bullet-comment timestamps, assets.js for capturing browser evidence with highlighted annotations, and watch.js for reading local video files through frame-sampling grids and speech-to-text.

Every stage of video development runs as a standard Claude Code skill defined inside directories containing SKILL.md files. Factory skills stored in defaults/skills/ are copied on first run to .claude/skills/ and never committed to source control. These encompass operational pipeline skills such as curate-topics, account-ideas, research-topic, write-script, precheck-content, make-video, revise-video, and review-data, alongside content production recipes like explainer-video and stage-montage, plus shared utility tools like voice-over and transcribe. Video rendering connects directly to the Apache-2.0 licensed HyperFrames engine. For animated narratives, SoloMCN integrates huashu-art-motion, using Claude to compose visual SVG assets or generate entire character expression sheets from a single source image.

What improved in SoloMCN – Claude Code as the runtime

Traditional AI video frameworks require rigid visual workflows where developers drag connection lines, set static prompt templates across dozen of individual nodes, and manage multiple per-tool API keys. When an API quota exhausts or an unhandled branch occurs, classic node graphs halt execution. SoloMCN eliminates fixed node pipelines by translating operational stages into natural-language skill definitions executed directly by Claude Code. When facing environmental hurdles during runtime, Claude detects the issue and executes remedial steps autonomously. During the recorded demonstration build, when cloud voice credits hit zero, Claude automatically installed a local offline speech synthesis engine, assigned three distinct local character voices across animal figures, and finalized the video without human intervention. When asked to adjust the voiceover style, it kept visual timelines intact and re-rendered audio and subtitles within ten minutes.

Autonomous research execution also received major upgrades over surface scrapers. During technical verification tests, Claude writes real code, executes command-line binaries, and records its own screen to yield factual visual footage. In a benchmark decompilation scenario, Claude authored a 79-line C game, compiled it with -O2 optimization flags, stripped debugging symbols, and tested an AI decompiler. Facing an architecture mismatch where local Ghidra lacked Apple silicon support, Claude compiled a compatible binary on the fly, reconstructed compilable C code in 96 seconds, and confirmed identical execution across 2,000 randomized inputs and 3,000 matches with matched 125-point outcomes.

What you gain from SoloMCN – Claude Code as the runtime

Deploying SoloMCN gives operators an autonomous production pipeline that spans from international trend detection to multi-platform publishing. Creators bypass developer fees and rate caps on social networks through a local browser cookie pool in the channels configuration. Users specify custom search syntax, such as pairings of artificial intelligence terms with education or employment themes, alongside minimum engagement thresholds. SoloMCN evaluates hourly velocity over 24-hour, 48-hour, and 7-day windows, feeding high-performing items into the topic radar. On YouTube, entering a free Data API key pulls top-viewed items, allowing Claude to parse captions, descriptions, and user discussions to generate original scripts targeted at new demographics.

The system enforces strict editorial control by restricting human intervention to three explicit checkpoints: approving the initial topic selection, reviewing the final rendered cut, and authorizing multi-platform distribution. Videos can be published automatically or set to manual entry mode across Douyin, Xiaohongshu, Bilibili, YouTube, and X using local authenticated Chrome sessions. For X, the system prepares bilingual copy adjusted to platform algorithms; for Douyin, it automatically retrieves relevant hashtags and campaign topics. Users can instruct the workbench to incorporate new production patterns in plain language, prompting Claude to draft line-by-line diffs for human confirmation before any skill modifications become permanent.

How to get SoloMCN – Claude Code as the runtime

Setting up SoloMCN requires a Mac running macOS with an active Claude subscription. The automated installation script inspects the host system, uses Homebrew to install Node.js version 22.9 or higher, Python 3.11 or higher, ffmpeg, Google Chrome, and Claude Code, clones the repository into ~/SoloMCN, and launches the web interface at port 5178.

curl -fsSL https://raw.githubusercontent.com/ymybxx/SoloMCN/main/install.sh | bash

Once installed, operators control the background daemon across any terminal directory using the dedicated solomcn command-line utility.

solomcn start
solomcn status
solomcn logs
solomcn stop

Manual deployment is supported by checking out the source tree, running setup scripts to prepare Python environments and HyperFrames rendering, and starting the local Node server.

git clone https://github.com/ymybxx/SoloMCN.git
cd SoloMCN
npm run setup
npm start

Claude Code models and credentials configure through the local environment or the settings panel. Users generate persistent authentication tokens through the Claude CLI to ensure uninterrupted background execution, and adjust model tiers between Opus, Sonnet, and Haiku directly within the workbench interface.

npm install -g @anthropic-ai/claude-code
claude
claude setup-token

To run individual pipeline steps manually inside the Claude Code interface without opening the web browser, users invoke standard slash commands pointing directly to the managed local skill definitions.

/curate-topics
/research-topic
/write-script
/make-video

What to watch after SoloMCN – Claude Code as the runtime

The open-source project roadmap outlines a full shift toward discrete autonomous AI staffing configurations. Upcoming milestones aim to formalize each production station into distinct configurable employee personas, including asset researchers, topic editors, screenwriters, video editors, and channel operators. Each role will maintain independent job specifications, targeted research directives, and isolated operational budgets, coordinating handoffs through structured internal tickets while operators manage overarching creative direction. Future development cycles will also implement fully automated daily topic generation passes that run on cron schedules to stage drafts ahead of human morning review.

Immediate architectural extensions focus on data feedback loops and expanded platform integrations. Planned updates will automate metric ingestion across published clips to review retention figures, comment sentiment, and view counts, cycling operational insights back into the curate-topics and write-script skill routines. Platform coverage will broaden beyond Douyin, Xiaohongshu, Bilibili, YouTube, and X to incorporate Kuaishou, WeChat Video Channels, and TikTok. Users must note existing operational constraints: cookie-based account pools violate X terms of service and risk throttling or bans, requiring secondary burner accounts, while automated browser publishing relies on current interface layouts that may require maintenance as platforms update their defensive bot protections.