diffchunk

MCP server that enables LLMs to navigate large diff files efficiently. Instead of reading entire diffs sequentially, LLMs can jump directly to relevant changes using pattern-based navigation.

Problem

Large diffs exceed LLM context limits and waste tokens on irrelevant changes. A 50k+ line diff can't be processed directly and manual splitting loses file relationships.

Solution

MCP server with 5 navigation tools:

• load_diff - Parse diff file with custom settings (optional)
• list_chunks - Show chunk overview with file mappings and per-file line counts (auto-loads)
• get_chunk - Retrieve specific chunk content (auto-loads)
• find_chunks_for_files - Locate chunks by file patterns (auto-loads)
• get_file_diff - Extract the complete diff for a single file (auto-loads)

How It Works

When you ask your AI assistant to analyze changes, it uses diffchunk's tools strategically:

1. Creates the diff file (e.g., git diff main..develop > /tmp/changes.diff) based on your question
2. Uses list_chunks to get an overview of the diff structure and total scope, including per-file line counts via file_details
3. Uses find_chunks_for_files to locate relevant sections when you ask about specific file types
4. Uses get_file_diff to fetch the complete diff for one specific file without loading an entire chunk
5. Uses get_chunk to examine specific sections without loading the entire diff into context
6. Tracks progress systematically through large changesets, analyzing chunk by chunk
7. Cleans up temporary files after completing the analysis

This lets your AI assistant handle massive diffs that would normally crash other tools, while providing thorough analysis without losing context.

Tool Usage Patterns

Overview first:

list_chunks("/tmp/changes.diff")
# -> 5 chunks across 12 files, 3,847 total lines, ~15,420 tokens
# Each chunk includes token_count and file_details with per-file line counts
# Response includes total_token_count for context-budget planning

Target specific files:

find_chunks_for_files("/tmp/changes.diff", "*.py")
# → [1, 3, 5] - Python file chunks

get_chunk("/tmp/changes.diff", 1)
# → Content of first Python chunk

Single-file diff:

get_file_diff("/tmp/changes.diff", "src/main.py")
# → Complete diff for src/main.py (header + all hunks)

# Glob patterns work when they match exactly one file
get_file_diff("/tmp/changes.diff", "*.config")
# → Complete diff for the single matching config file

Systematic analysis:

# Process each chunk in sequence
get_chunk("/tmp/changes.diff", 1)
get_chunk("/tmp/changes.diff", 2)
# ... continue through all chunks

Supported Formats

• Git diff output (git diff, git show)
• Unified diff format (diff -u)
• Multiple files in single diff
• Binary file change indicators

Performance

• Efficiently handles 100k+ line diffs
• Memory efficient streaming
• Auto-reload on file changes

Documentation

• Design - Architecture and implementation details
• Contributing - Contributing guidelines and development setup

Setup

Prerequisite: Install uv (an extremely fast Python package manager) which provides the uvx command.

Add to your MCP client configuration:

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

Usage

Your AI assistant can now handle massive changesets that previously caused failures in Cline, Roocode, Cursor, and other tools.

Using with AI Assistant

Once configured, your AI assistant can analyze large commits, branches, or diffs using diffchunk.

Here are some example use cases:

Branch comparisons:

• "Review all changes in develop not in the main branch for any bugs"
• "Tell me about all the changes I have yet to merge"
• "What new features were added to the staging branch?"
• "Summarize all changes to this repo in the last 2 weeks"

Code review:

• "Use diffchunk to check my feature branch for security vulnerabilities"
• "Use diffchunk to find any breaking changes before I merge to production"
• "Use diffchunk to review this large refactor for potential issues"

Change analysis:

• "Use diffchunk to show me all database migrations that need to be run"
• "Use diffchunk to find what API changes might affect our mobile app"
• "Use diffchunk to analyze all new dependencies added recently"

Direct file analysis:

• "Use diffchunk to analyze the diff at /tmp/changes.diff and find any bugs"
• "Create a diff of my uncommitted changes and review it"
• "Compare my local branch with origin and highlight conflicts"

Tip: AI Assistant Rules

Add to your AI assistant's custom instructions for automatic usage:

When reviewing large changesets or git commits, use diffchunk to handle large diff files.
Create temporary diff files and tracking files as needed and clean up after analysis.

Configuration

Path Requirements

• Absolute paths only: /home/user/project/changes.diff
• Cross-platform: Windows (C:\path) and Unix (/path)
• Home expansion: ~/project/changes.diff

Auto-Loading Defaults

Tools auto-load with optimized settings:

• max_chunk_lines: 1000
• skip_trivial: true (whitespace-only)
• skip_generated: true (lock files, build artifacts)

Custom Settings

Use load_diff for non-default behavior:

load_diff(
    "/tmp/large.diff",
    max_chunk_lines=2000,
    include_patterns="*.py,*.js",
    exclude_patterns="*test*",
    context_lines=2
)

Format Options

Use the format parameter on get_chunk to transform output for LLM consumption:

# Default - raw diff output
get_chunk("/tmp/changes.diff", 1, format="raw")

# Annotated - structured with line numbers, file headers, hunk separation
get_chunk("/tmp/changes.diff", 1, format="annotated")

# Compact - token-efficient, only new hunks (context + added lines)
get_chunk("/tmp/changes.diff", 1, format="compact")

Annotated format adds ## File: headers, __new hunk__/__old hunk__ sections with new-file line numbers, and function context from @@ headers.

Compact format shows only what was added or kept, omitting removed lines and __old hunk__ sections entirely. Useful when you only need to see the final state.

Context Reduction

Use context_lines on load_diff to reduce context lines per hunk at load time:

# Keep only 2 lines of context around each change
load_diff("/tmp/large.diff", context_lines=2)

# Keep only changes, no context
load_diff("/tmp/large.diff", context_lines=0)

This composes with format - context is reduced at load time, then formatting is applied at display time.

License

MIT