By Kodelyth · updated · v2.30.0
Kodelyth ECC vs Plain Claude Code
Short version. Plain Claude Code is a complete, polished AI coding environment. Kodelyth ECC is a free, MIT-licensed toolkit that layers 71 specialist AI agents, a universal MCP server, local memory recall, and two independent token-compression layers on top — without replacing anything Claude Code already does. If you're new to Claude Code, start with it alone. If you've hit any of the symptoms below, add ECC.
When plain Claude Code is enough
- You work in one AI tool and don't need the same agents in Cursor or Windsurf.
- Your token bill is comfortable and you haven't noticed input/output costs.
- You don't find yourself typing the same multi-step prompts repeatedly ("review this for security, performance, and a11y").
- You don't need an MCP server connecting your tools to Claude Desktop or LangGraph.
- You rarely ask structural questions across many files ("who calls this?").
There is no shame in staying here. Claude Code alone is a strong setup.
When ECC starts earning its keep
- "Which agent handles X?" is friction → intent routing picks one of 71 specialists from plain language. No names to remember.
- Your token bill is annoying → RTK compresses shell output 60-90% before the model sees it; Terse compresses replies 40-70%. 55-65% combined on a typical session.
- You keep saying "review this for security, perf, and a11y" →
/team-reviewfires 4 specialists in parallel. Same result, one line. - "Who calls this function?" takes 20 file reads → the codebase graph answers it in one MCP call. ~99% fewer tokens.
- You rediscover the same fix every few weeks → Memory Core recalls past solutions on every new prompt so you don't pay twice.
- You want the same agents in Cursor, Codex, or Kiro tomorrow → ECC installs the same catalog across 12 IDEs in one command.
What's in the box (plain Claude Code vs with ECC)
| Capability | Plain Claude Code | + Kodelyth ECC |
|---|---|---|
| The core AI coding environment | ✓ | ✓ |
Project-level CLAUDE.md rules | ✓ | ✓ |
| Hooks system | ✓ | ✓ (ECC adds 44 of its own) |
| MCP server support (consumer) | ✓ | ✓ |
| Specialist agents routable by intent | — | ✓ 71 of them |
| Universal MCP server (provider) | — | ✓ Auto-wires 6 IDEs |
| Parallel multi-agent commands | — | ✓ 8 of them |
| Shell-output token compression (RTK) | — | ✓ 60-90% input reduction |
| AI-reply token compression (Terse) | — | ✓ 40-70% output reduction |
| Codebase AST graph (158 languages) | — | ✓ |
| Local memory that survives repo moves | — | ✓ Memory Core |
| Autonomous orchestrator for multi-step objectives | — | ✓ Super Agent |
| Localhost observability dashboard | — | ✓ |
The "✓ in both" rows are exactly what you'd expect — ECC does not patch or override Claude Code's core behavior. The additions are additive.
How to install
npm i -g kodelyth-ecc
kodelythecc --target claude-code --codebase-graphRestart Claude Code. That's the entire install. ECC drops its agents, skills,
commands, hooks, and rules into ~/.claude/ and auto-registers its MCP server.
Token savings (RTK + Terse) are wired at the same time.
How to invoke it
You don't type agent names. Just say what you're trying to do:
"I've been staring at this NullPointerException for two hours."
→ Routes to debug-detective
"Review this before I merge."
→ Routes to code-reviewer (or language-specific reviewer by file extension)
"Something is slow and I can't tell what."
→ Routes to performance-optimizer
"Index this project."
→ codebase-memory-mcp builds the AST graphFor the parallel commands, type them:
/team-review # code-reviewer + security-reviewer + perf + api-guardian
/security-audit # security + dependency + api-guardian
/devil-mode # prompt-injection + supply-chain + secret + backdoor hunters
/debug-blitz # debug + silent-failure + env-debugger (for stuck bugs)What stays the same
Your Claude Code config, your CLAUDE.md, your hooks, your MCP servers — all
untouched. ECC adds alongside; it never overwrites. If you uninstall:
kodelythecc uninstall --dry-run # see what will be removed
kodelythecc uninstall --yes # actually remove
kodelythecc uninstall --yes --keep-memory # keep the local memory storeNothing else on your machine is touched.
The honest tradeoff
ECC is more moving parts. More files in ~/.claude/. Another binary on PATH.
If you don't need any of the capabilities above, those moving parts are cost
without benefit. If you need even two of them, the install pays for itself in
the first session.
Bottom line
Plain Claude Code is great alone. ECC adds the specialist agents, the token savings, the local memory, and the MCP bridge — without replacing anything. 2-minute install, zero telemetry, MIT licensed. If any of the symptoms up top describe you, add it; if none do, stay where you are.
npm i -g kodelyth-ecc && kodelythecc --target claude-code --codebase-graph