MCP Context Manager MCP Server

MCP Context Manager

The smartest code search for Claude. Reduces token usage by 70-90% with BM25 ranking, fuzzy matching, and natural language queries.

What is this?

An MCP (Model Context Protocol) server that gives Claude superhuman code navigation. Instead of reading entire files (thousands of tokens), Claude queries for exactly what it needs (tens of tokens).

The difference:

- Claude reads 3 files → 5,400 tokens → 5 seconds
+ Claude queries "auth middleware" → 230 tokens → 85ms

95% token savings + 10x faster = drastically lower API costs.

Features

• 🎯 Natural Language Search - "authentication middleware" finds all auth code
• 🔍 BM25 Ranking - Industry-standard relevance algorithm (like Elasticsearch)
• ✨ Fuzzy Matching - Handles typos automatically ("athenticate" → "authenticate")
• 🧠 Smart Tokenization - Understands camelCase, snake_case, paths
• 🚀 Blazing Fast - <100ms searches on 10,000+ file repos
• 🌐 Multi-Language - TypeScript, JavaScript, Python, Go, Rust, Java, C/C++, C#, Lua
• 💰 Token Savings - 70-90% reduction in typical usage

Quick Start

Installation

npm install -g claude-mcp-context

Automatic Setup (Claude Code CLI):

The package automatically detects and configures Claude Code. After installation, run:

mcp-context-setup

This will:

1. ✅ Build the server (if needed)
2. ✅ Verify server functionality
3. ✅ Register with Claude Code automatically
4. ✅ Confirm registration

Manual Configuration (Claude Desktop or troubleshooting):

If automatic setup doesn't work or you're using Claude Desktop, add to your config file:

Claude Desktop (~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "context-manager": {
      "command": "claude-mcp-context"
    }
  }
}

Claude Code (~/.claude/mcp.json):

{
  "mcpServers": {
    "context-manager": {
      "command": "claude-mcp-context"
    }
  }
}

Verify Installation:

claude mcp list  # Should show "context-manager"

Usage

Restart Claude and try:

"Index this repository"
"Find the authentication middleware"
"Show me all payment-related functions"
"What does the UserService class do?"

That's it! Claude will now use intelligent search instead of reading entire files.

How It Works

1. Index Your Repository (One Time)

You: "Index this repository"

MCP Context Manager parses all code and builds a searchable index (~1 second per 1000 files).

2. Natural Language Queries

You: "Find authentication middleware"

Returns:
📍 authMiddleware (function) - Score: 892.3
   src/middleware/auth.ts:15

📍 AuthService (class) - Score: 654.2
   src/services/AuthService.ts:8

3. Massive Token Savings

Instead of 5,400 tokens to read 3 files, you get exactly what you need in 230 tokens.

Search Intelligence

Multi-Word Queries

"payment validation" → finds validatePayment(), PaymentValidator, checkPayment()
"user auth service" → finds UserAuthService, authenticateUser(), etc.

Fuzzy Matching

"athenticate" → suggests authenticate
"usrService" → suggests userService

Query Expansion

"auth" → searches: auth, authenticate, authentication, authorization
"db" → searches: db, database, data
"config" → searches: config, configuration, configure, settings

Code Body Search

Searches inside function implementations, not just names:

"jwt token" → finds code that uses JWT, even if function isn't named jwt*

Available Tools

Claude automatically chooses the best tool based on your query.

Examples

Find Authentication Code

You: "Find all authentication-related code"

Claude uses: get_relevant_context("authentication")

Returns:
- authMiddleware() in middleware/auth.ts
- authenticate() in auth/handler.ts
- AuthService class in services/auth.ts
- validateToken() in utils/jwt.ts

Tokens: 340 (vs 4,200 reading all files) = 92% savings

Debug Payment Flow

You: "Show me the payment processing flow"

Claude uses: get_rel