Largefile MCP Server

Navigate, search, and edit large codebases, logs, and data files that exceed AI context limits.

Why Largefile?

• Go beyond context limits - Read, search, and edit files too large to fit in AI context windows
• Semantic code navigation - Tree-sitter extracts functions/classes for Python, JS/TS, Rust, Go
• Fewer LLM errors - Search/replace editing eliminates line number mistakes common with line-based edits
• Smart search - Fuzzy matching, regex, case-insensitive, inverted, and count-only modes
• No size limits - Handles multi-GB files via tiered memory strategy (RAM → mmap → streaming)

Quick Start

Prerequisite: Install uv for the uvx command.

{
  "mcpServers": {
    "largefile": {
      "command": "uvx",
      "args": ["--from", "largefile", "largefile-mcp"]
    }
  }
}

Tools

Tool	Use For
get_overview	File structure and semantic outline before diving in
search_content	Finding patterns, counting occurrences, regex matching
read_content	Reading specific sections; tail/head modes for logs
edit_content	Safe search/replace with automatic backups
revert_edit	Recovering from bad edits
list_directory	Browse directory trees with recursive depth control
search_directory	Search patterns across all files in a directory

When to Use Largefile

Use when:

• File exceeds ~1000 lines or 100KB (supports multi-GB files)
• Navigating large codebases with semantic structure
• Analyzing log files (especially recent entries with tail mode)
• Making search/replace edits across large files
• Counting occurrences without loading full content

Don't use for:

• Small files that fit in context (AI doesn't need help with those)
• Binary files (images, executables, compressed)

Usage Examples

Large Codebase Navigation

# Get semantic structure of a large Python file
overview = get_overview("/path/to/large_module.py")
# Returns: 2,847 lines, 15 classes, function outline via Tree-sitter

# Find all class definitions
classes = search_content("/path/to/large_module.py", "class ", fuzzy=False)

# Read complete class with semantic chunking
code = read_content("/path/to/large_module.py", pattern="class UserModel", mode="semantic")

Batch Refactoring

# Preview rename across file
preview = edit_content("/path/to/api.py", changes=[
    {"search": "process_data", "replace": "transform_data"},
    {"search": "old_endpoint", "replace": "new_endpoint"}
], preview=True)

# Apply changes (creates automatic backup)
result = edit_content("/path/to/api.py", changes=[...], preview=False)

# Undo if needed
revert_edit("/path/to/api.py")

Log Analysis

# Get log file overview
overview = get_overview("/var/log/app.log")
# Returns: 150,000 lines, 2.1GB

# Read last 500 lines efficiently
recent = read_content("/var/log/app.log", limit=500, mode="tail")

# Count errors without loading content
error_count = search_content("/var/log/app.log", "ERROR", count_only=True, fuzzy=False)

# Find errors with regex
errors = search_content("/var/log/app.log", r"ERROR.*timeout", regex=True)

Supported Languages

Tree-sitter semantic analysis for: Python, JavaScript/JSX, TypeScript/TSX, Rust, Go, Java

Other file types use text-based analysis with graceful fallback.

File Size Handling

Size	Strategy
< 50MB	Full memory loading with AST caching
50-500MB	Memory-mapped access
> 500MB	Streaming (tail/head modes recommended)

Configuration

Environment variables for tuning:

LARGEFILE_MEMORY_THRESHOLD_MB=50      # RAM loading limit
LARGEFILE_MMAP_THRESHOLD_MB=500       # Memory mapping limit
LARGEFILE_FUZZY_THRESHOLD=0.8         # Match sensitivity (0.0-1.0)
LARGEFILE_MAX_SEARCH_RESULTS=20       # Results per search
LARGEFILE_BACKUP_DIR=~/.largefile/backups

Documentation

• API Reference - Detailed tool documentation
• Configuration Guide - All environment variables
• Examples - More workflow examples
• Design Document - Architecture details
• Contributing - Development setup