Claude Code + Entroly MCP Setup

Last updated: July 2026 | Reading time: 5 min

Claude Code is one of the most powerful AI coding tools available — Anthropic's Claude model is exceptional at reasoning about code. But even Claude Code has a critical blind spot: it can only see a fraction of your codebase at a time.

When Claude Code does not have the right evidence, it can invent function signatures, miss config files, or break dependencies it did not inspect. This guide shows the clean setup: verify Entroly locally, then add it to Claude Code through MCP.

Why Claude Code Hallucinates

Claude Code uses file search, grep, and open files to build context. On a large project, it faces a fundamental constraint: the context window has a token limit. Claude Code is smart about what it reads, but it can't read everything.

The result:

• Partial visibility — Claude Code sees files relevant to your query, but misses related configs, utilities, and type definitions
• Wasted tokens — when it does read files, it reads them in full — including boilerplate, comments, and duplicate code
• No dependency awareness — it doesn't know that auth.py depends on auth_config.py unless it explicitly searches for it
• No learning — each conversation starts fresh with no memory of which context worked best

The Fix: Entroly as an MCP Server

Entroly plugs into Claude Code as an MCP (Model Context Protocol) server. Instead of Claude Code searching for files manually, Entroly delivers the optimal context automatically:

Setup

Step 1: Install and verify locally

pip install -U entroly
entroly verify-claims
entroly simulate

Step 2: Connect to Claude Code

claude mcp add entroly -- entroly

Claude Code stays your client. Entroly adds local MCP tools for context optimization, exact retrieval, receipts, feedback, and savings reports. You do not need proxy mode for this subscription workflow.

Step 3: Inspect value anytime

entroly simulate
entroly status
entroly doctor

Use these commands to inspect local savings estimates, runtime status, and common setup issues without making an LLM call.

What Happens Behind the Scenes

When you ask Claude Code a question, Entroly:

1. Analyzes your query — extracts intent, keywords, and relevant code areas
2. Selects optimal context — uses a knapsack solver to pick the best fragments that fit the token budget
3. Delivers at 3 resolution levels — critical files in full, supporting files as signatures, peripheral files as references
4. Tracks what worked — reinforcement learning improves selection over time

The goal is better evidence with fewer input tokens. Exact latency and savings depend on repo size, installed engine, query, and budget.

Advanced: Claude Code + Entroly Pro Tips

Use quality presets for different tasks

# MCP path for Claude Code subscriptions
claude mcp add entroly -- entroly

# Proxy path only when you control a provider API key
entroly proxy --quality balanced

Check your savings

entroly dashboard

The dashboard shows lifetime token savings, cost trends, codebase health grades, and PRISM weight evolution.

Role-based optimization

entroly role backend   # optimizes for backend-heavy context
entroly role frontend  # optimizes for UI/component context
entroly role fullstack # balanced

FAQ

Does my code leave my machine?

No. Entroly runs 100% locally. It processes your code on your machine and delivers context to Claude Code locally via MCP.

Does it work with Claude Code's free usage?

Yes. Every tier benefits — fewer tokens per request means more requests within your quota.

Can I use it alongside CLAUDE.md files?

Absolutely. CLAUDE.md gives Claude Code your project instructions. Entroly gives it your code context. They're complementary.

pip install -U entroly
entroly verify-claims
entroly simulate
claude mcp add entroly -- entroly