mgrep-code-search

mgrep serves as a semantic search engine specifically calibrated for source code and documentation. Unlike traditional text-matching utilities, it interprets the intent of your search queries to identify relevant code segments, logic patterns, or architectural implementations across directories. By indexing files locally, it allows for fast, context-aware discovery in repositories that are too large for manual review. It supports multi-modal search, covering text, PDFs, and images, ensuring that developers can locate hidden definitions or scattered implementations without knowing exact keywords. By running a local watcher, the tool maintains a fresh index that adheres to project-specific exclusion rules like .gitignore. This provides an effective way to navigate legacy systems or unfamiliar codebases where the mental map of the structure has not yet been formed.

When to Use This Skill

• •Locating the specific implementation of an authentication flow
• •Finding scattered patterns of error handling across microservices
• •Identifying all files related to a feature described in plain English
• •Exploring an unfamiliar codebase structure for the first time

How to Invoke This Skill

Example prompts that trigger this skill in Claude Code, Cursor, or Antigravity:

• “Where is the user session logic handled?
• “Show me the code that validates API request bodies
• “How does the database connection initialization work?
• “Find patterns for logging inside the payment services directory
• “What defines the current UI theme configuration?

Pro Tips

• 💡Combine semantic search with traditional `grep`/`ripgrep` for a hybrid approach: use `mgrep` for conceptual understanding and `grep` for precise string matches.
• 💡Ensure the `mgrep watch` command is running in your development environment to keep your codebase index up-to-date with file changes.
• 💡Refine natural language queries by starting broad and gradually narrowing down the scope to achieve more precise and relevant results.

What this skill does

• •Indexes and maintains searchable code vectors locally
• •Performs semantic retrieval based on natural language intent
• •Generates AI-powered summaries of search results for direct answers
• •Supports scoping searches to specific directory paths
• •Configurable result ranking and output formatting via CLI flags

When not to use it

• ✕Searching for precise literal string matches or specific variable names
• ✕Performing regex-based bulk refactoring operations
• ✕Operating in environments without the Bun runtime

Example workflow

1. Install and confirm the Bun runtime is available
2. Run 'bunx @mixedbread/mgrep watch' in the root directory to build the index
3. Execute a natural language query using 'bunx @mixedbread/mgrep "search query"'
4. Refine the scope by adding a path argument like 'src/components'
5. Use the '-a' flag to generate a synthetic AI summary of the findings

Prerequisites

• –Bun runtime installed on the system
• –Existing project directory to index

Pitfalls & limitations

• !Requires active background watcher to keep index synchronized with live file changes
• !Not suitable for strict exact-match searches where grep is faster
• !Index bloat in extremely large repositories might affect local performance

FAQ

How it compares

While standard grep requires specific keyword knowledge, mgrep maps abstract concepts to code snippets, acting as a semantic bridge that eliminates the 'what word did I name this function' struggle.

Source & trust

Recommended Rules

Recommended Workflows

Recommended MCP Servers