The Claude Code Ecosystem: Everything You Can Plug In

You use Claude Code every day. So do roughly 370 new developers each week, judging by the repos they create. PT-Edge tracks 2,400+ repositories in the Claude Code ecosystem, scored daily on quality, maintenance, and adoption. Most of them are noise — the average repo in claude-skill-orchestration has 15 stars and zero releases. This guide cuts through it. If you already live in the terminal with Claude Code, here is what is actually worth installing, what to watch, and what to ignore.

The ecosystem at a glance: 2,400 repos, 90% noise

Before you install anything, understand the scale and the signal-to-noise problem. The Claude Code ecosystem is growing at roughly 370 new repos per week. Most of them are forks, clones of clones, or weekend experiments that never ship a second commit.

Subcategory	Repos	Avg Stars	Max Stars	Signal Level
Team Systems	241	433	74,013	Strong
Skills	161	339	43,585	Strong
Development Templates	299	85	7,779	Moderate
Configuration	225	176	3,588	Moderate
Platforms	187	241	41,385	Moderate
Workspaces	162	135	6,764	Moderate
Skill Orchestration	300	15	899	Mostly noise
Plugins	113	30	1,602	Mostly noise

The first column that matters is "Avg Stars." Anything below 100 is a category full of abandoned experiments. claude-code-team-systems (241 repos, avg 433 stars) and claude-code-skills (161 repos, avg 339 stars) are where the actual tools with traction live. The skill-orchestration subcategory with 300 repos and a 15-star average is the noise floor personified.

The hottest category right now is claude-code-team-systems with a 2.21x acceleration rate and 2,667 new stars in the last 7 days alone. Multi-agent orchestration is where the energy is. Memory (1.90x acceleration) is close behind.

Memory and persistence: the first problem to solve

Claude forgets between sessions. Every power user hits this within their first week, and it is the most common complaint on Hacker News threads. The ecosystem has responded with at least six distinct approaches to solving it. None has become a standard, but one has clear momentum.

claude-mem (87/100, 34,460 stars, 7,318 downloads/month) dominates. It automatically captures everything Claude does during sessions and compresses it using Claude's own agent-sdk. With 8 releases in the last 30 days and 82 commits, it is the memory solution with the most sustained investment. Install this first.

But claude-mem is not the only approach worth knowing about:

Approach	Project	Score	Stars	How it works
Agent-SDK compression	claude-mem	87/100	34,460	Auto-captures session data, AI-compresses it, persists across sessions
Hooks + Markdown	claude-memory-engine	—	59	Zero dependencies. Git hooks write to markdown files. Simplest possible approach
Knowledge graph	code-review-graph	—	1,144	Local knowledge graph of your codebase. Claims 6.8x fewer tokens on reviews
Portable .mv2 files	claude-brain	—	322	Single portable file. No database, no SQLite, no ChromaDB
MCP-based	cipher	67/100	3,578	Open-source memory layer designed specifically for coding agents
Tiered write-gate	total-recall	—	185	Correction propagation and write gates to prevent memory pollution

Anthropic also shipped AutoDream memory consolidation in late March — a background process modeled on REM sleep that manages memory without explicit user intervention. This is the first-party answer, and it may make some of these third-party solutions redundant over time. But today, claude-mem is the most practical choice. For a deeper breakdown of memory architectures and how they compare across agent frameworks, see our agent memory landscape deep dive.

Skills that actually matter

The curated lists are the entry point. awesome-claude-skills (45/100, 43,585 stars) from Composio is the most popular. awesome-claude-code (27,790 stars) is the broadest, covering skills, hooks, slash-commands, and agent orchestrators. claude-skills (61/100, 6,484 stars) offers 66 specialized skills for full-stack developers — the most opinionated and immediately useful of the three.

But there is a crucial distinction most people miss: skills, hooks, and slash commands are not the same thing.

Skills are markdown instruction files that live in .claude/skills/. They shape how Claude approaches specific tasks — "when asked to write tests, use this pattern." They are passive context, not executable code.
Hooks are shell scripts triggered by lifecycle events (pre-tool, post-tool, notification). They execute real code: linting before commits, logging after tool calls, auto-formatting on save. This is where memory plugins like claude-memory-engine operate.
Slash commands are user-invokable shortcuts defined in .claude/commands/. They bundle a prompt template you trigger with /command-name.

Most "skill collections" mix all three without distinguishing them, which is why installing a random skills repo often does nothing useful. The 300-repo claude-skill-orchestration subcategory (average 15 stars) is full of this confusion.

Top picks by use case

Use case	Install this	Stars	Why
Full-stack dev	claude-skills	6,484	66 focused skills for React, Node, databases, testing, deployment
Team orchestration	everything-claude-code	74,013	The complete harness: skills, instincts, memory, security. 577 commits/30d
Research / academic	claude-scholar	1,341	Personal CLI config for academic research and software development
DevOps / automation	claude-code-tools	1,577	Practical productivity tools for CLI coding agents
Game dev (Godot)	godogen	1,070	Builds complete Godot 4 projects from a game description

For the broader architecture of how skills compose into agent workflows, see our agent skills architecture deep dive.

The MCP stack: your shopping list

MCP servers extend what Claude Code can perceive and act on. The ecosystem has hundreds, but the practical stack for daily use comes down to five.

Server	Score	Stars	Downloads/mo	What it does
chrome-devtools-mcp	87/100	28,511	2,689,135	Browser debugging and automation from the terminal
serena	69/100	21,348	0	Semantic code retrieval and editing across your codebase
n8n-mcp	91/100	14,908	405,542	Build and trigger n8n automation workflows
firecrawl-mcp-server	64/100	5,738	0	Web scraping and content extraction
claude-context	64/100	5,625	0	Full codebase search with semantic understanding

Chrome DevTools MCP is the essential. At 28,511 stars and 2,626,019 downloads per month, it is the most-installed MCP server in the Claude Code orbit by a factor of 10. It gives Claude Code eyes into the browser — inspect elements, read console logs, debug rendering issues, all without leaving the terminal. If you install one MCP server, make it this one.

Serena (69/100, 21,348 stars) is the second priority. It gives Claude Code semantic understanding of your codebase — not just text search but structural retrieval that understands functions, classes, and relationships. With 188 commits in the last 30 days, it is actively maintained and improving.

n8n-mcp (14,908 stars, 257,296 downloads/month) bridges Claude Code to workflow automation. If you use n8n for CI/CD, data pipelines, or operational workflows, this lets Claude build and trigger them directly. It had 26 releases in the last 30 days — the most actively shipped MCP server in the ecosystem.

For deeper coverage of browser automation and perception tools, see our browser automation deep dive.

Multi-session orchestration: running a fleet of Claudes

Once you are productive with a single Claude Code session, the next question is: can I run five in parallel? The answer is yes, and two patterns have emerged.

Pattern 1: tmux + git worktrees

The lightweight approach. You spin up multiple Claude Code instances in tmux panes, each in its own git worktree, each working on a different branch or task. ccpm (47/100, 7,626 stars) is the most structured version — it uses GitHub Issues as a task queue and git worktrees for isolation. agent-of-empires (1,103 stars, 0 commits/30d) manages sessions across 8+ agent CLIs via tmux.

Pattern 2: dedicated TUI

Purpose-built terminal UIs for managing multiple agent sessions.

Project	Score	Stars	Downloads/mo	Approach
ccmanager	76/100	946	3,995	Session manager for 8+ agent CLIs. The only one with real downloads
Kaku	63/100	3,045	0	Fast Rust terminal built for AI coding. 350 commits/30d
crystal	53/100	2,980	0	Now Nimbalyst. Parallel sessions in git worktrees

ccmanager is the practical choice here. At 3,779 downloads per month, it is the only session manager with meaningful adoption beyond GitHub stars. It supports Claude Code, Gemini CLI, Codex CLI, Cursor Agent, and Copilot CLI from a single interface.

For teams, the picture is different. everything-claude-code (74,013 stars, 577 commits/30d) is the full performance optimization harness. awesome-claude-code-subagents (13,597 stars) provides 100+ pre-built specialist subagents you can delegate to. oh-my-claudecode (65/100, 9,561 stars) is the most actively developed multi-agent tool, with 554 commits in 30 days — teams-first orchestration designed for dividing work across coordinated Claude Code instances.

Workspaces and platforms: wrapping Claude Code in a GUI

Not everyone wants a terminal-only experience. Several projects wrap Claude Code (and other agent CLIs) in a graphical interface.

Cherry Studio (71/100, 41,385 stars) is the leader — a full AI productivity studio with smart chat, autonomous agents, and 300+ assistants. It provides unified access to frontier LLMs including Claude, and with 215 commits in 30 days, it is a mature and actively maintained platform.

opcode (50/100, 20,905 stars) is the GUI toolkit specifically for Claude Code — create custom agents, manage interactive sessions, run secure background agents. superset (66/100, 6,764 stars, 397 commits/30d) bills itself as "IDE for the AI Agents Era" — run a fleet of Claude Code, Codex, and other agents from one interface. For containerised setups, HolyClaude (1,061 stars) packages Claude Code with a web UI, five AI CLIs, a headless browser, and 50+ tools in a single Dockerfile.

Configuration and governance: CLAUDE.md patterns

Your CLAUDE.md file is arguably the single most important file in your project when working with Claude Code. It shapes every response, every tool call, every refactor. Getting it right matters more than which skills you install.

claude-code-guide (68/100, 3,588 stars) is the best starting reference — setup, commands, workflows, agents, skills, and tips from beginner to power user. claude-bootstrap (51/100, 535 stars) takes a security-first, spec-driven approach to project initialization — opinionated defaults that prevent Claude from running dangerous commands or accessing files it should not touch.

For tracking what Claude Code itself is doing under the hood, claude-code-changelog (631 stars) is invaluable. With 5,791 commits, it tracks every release's prompts, feature flags, and metadata. When Claude Code's behavior changes overnight — and the "Claude Code is being dumbed down?" thread (1,085 HN points) shows how common this feeling is — this repo tells you exactly what changed.

For the broader picture of how teams govern AI coding agents in production, see our agent governance deep dive.

Computer Use: agents that can click

Anthropic shipped Computer Use integration for Claude Code on March 31, 2026. Agents can now open applications, click UI elements, test the apps they build, and interact with browsers — all from the CLI. This is the newest capability and the one most likely to reshape the MCP stack over the next few months.

The ecosystem was already building toward this. BrowserWing (1,065 stars) turns browser actions into MCP commands or Claude Skills. HolyClaude bundles a headless browser into the Claude Code workstation. But the first-party computer use integration changes the calculus: if Claude Code can natively control a browser, the third-party wrappers need to offer something beyond basic page interaction to survive.

What to watch: the intersection of Computer Use and MCP. When Claude Code can both build a web app and test it by clicking through the UI, the feedback loop tightens dramatically. The 7-day autonomous loops that Anthropic recently shipped become more practical when the agent can verify its own work visually. For the full picture on perception and browser automation, see our browser automation deep dive.

What is shipping versus what is vaporware

The hardest signal to fake is release velocity. Of the 2,400+ repos in the Claude Code ecosystem, only 9 had formal releases in the last 30 days. Here is who is actually shipping:

Project	Releases (30d)	Commits (30d)	Latest Release
pilot-shell	42	131	2026-04-01
n8n-mcp	26	40	2026-03-31
AionUi	19	1,330	2026-04-01
claude-mem	8	82	2026-03-29
Cherry Studio	8	215	2026-03-27
everything-claude-code	2	577	2026-03-21

pilot-shell at 42 releases in 30 days is shipping at a furious pace — roughly one release per day. n8n-mcp at 26 releases is the most actively maintained MCP server. After that, there is a steep drop. The vast majority of the 2,400 repos have zero formal releases, zero downloads, and minimal commit activity. Stars without releases is hype. Releases without downloads is a hobby project. Only a handful of projects in this ecosystem pass both filters.

The download reality check

Downloads tell the bluntest truth about adoption:

Chrome DevTools MCP: 2,626,019/month — genuine mass adoption
n8n-mcp: 257,296/month — strong workflow automation niche
zcf: 13,170/month — zero-config code flow finding its audience
claude-mem: 6,486/month — the memory standard, growing
ccmanager: 3,779/month — the session manager that people actually install
Then a cliff. Most repos in this ecosystem have exactly zero downloads.

The Hacker News temperature check

Claude Code generated 25 Hacker News posts exceeding 300 points in the last 90 days. The top post (1,957 points) was the source code leak via a map file in the npm registry. The themes reveal what the community is feeling:

Anxiety about quality: "Claude Code is being dumbed down?" (1,085 points, 702 comments) — the most engaged thread after the leak. Developers are worried about degradation and want objective benchmarks.
Vendor lock-in fears: "Anthropic blocks third-party use" (625 points), "I was banned for scaffolding a CLAUDE.md file" (749 points), "Developing a competitor using Claude Code is banned" (321 points) — a persistent undercurrent of concern about Anthropic's control.
Excitement about new capabilities: "Claude Code's new hidden feature: Swarms" (521 points), "Claude Code Remote Control" (544 points), multi-session orchestration (396 points) — the community is building faster than Anthropic can ship features.
Cultural adoption: "Tell HN: I'm 60 years old. Claude Code has re-ignited a passion" (344 points), "Claude Code is suddenly everywhere inside Microsoft" (409 points) — this has crossed the chasm from early adopters to mainstream engineering teams.

Every major AI newsletter — Latent Space, Simon Willison, Zvi, Pragmatic Engineer, One Useful Thing — covers Claude Code multiple times per month. The source code leak on March 31 revealed auto-dream memory, frustration regexes, and undercover mode. The coverage is overwhelming and shows no sign of slowing.

What to install right now

If you have been reading this far and want a concrete starting point, here is the minimum viable Claude Code stack:

Memory: claude-mem — install this first. Session persistence is the single biggest quality-of-life improvement.
MCP - browser: Chrome DevTools MCP — gives Claude Code eyes into the browser. Essential for web development.
MCP - code search: claude-context or Serena — semantic codebase understanding. Pick one based on whether you want search (claude-context) or editing (Serena).
Skills: claude-skills — 66 focused skills for full-stack development. More useful than the mega-lists.
Configuration: claude-code-guide — read it, steal the CLAUDE.md patterns, adapt them to your project.

Skip the orchestration tools until you have maxed out a single session. Skip the platforms unless you genuinely need a GUI. And ignore anything in the claude-skill-orchestration subcategory with fewer than 500 stars — 90% of it is noise.

How to use this data

Every project mentioned in this guide has a quality-scored page in our directory, updated daily. You can:

Browse all Claude Code categories to find projects in your specific niche
Check trending Claude Code projects to see what is gaining momentum this week
Explore the MCP server directory for tools that extend Claude Code's capabilities
Compare any two projects side by side on maintenance, adoption, maturity, and community scores

Quality scores update daily from live GitHub, PyPI, and npm data. When a project stops shipping releases, the score drops. When a project starts gaining real downloads, the score rises. In an ecosystem adding 370 repos per week, the data does the curation work so you do not have to manually audit each repo's commit history before trusting it with your codebase.