Unch is an MCP server tool developed by Uchebnick, built primarily with the Go programming language. This free program is categorized as an MCP server and is designed for semantic code search. It assists developers and engineers in discovering code symbols and documentation by understanding code behavior rather than relying on exact names. Unch can be installed via npm, Homebrew, a shell installer, PowerShell, or directly using "go install", providing flexible distribution options across various platforms.

What does Unch connect to?

Unch functions as a local-first semantic code search tool, connecting to code repositories to index and search code symbols and documentation. It adheres to the Model Context Protocol (MCP) standard, which implies support for stdio and Streamable HTTP transport mechanisms, with SSE also being a common type. Unch integrates with various AI clients, including Claude Code, Cursor, GitHub Copilot, Kiro, and OpenCode, with guides available for these platforms. Broader MCP-compatible clients often include Claude Desktop, Cursor AI, Windsurf, and LibreChat.

Core features and tools

The program exposes several callable tools to manage code search and integration. Key tools include index for repository scanning, search for querying code, and remote for managing external connections. Related operations or sub-commands provide additional functionality, such as auth openrouter for provider authentication, create ci and bind ci for continuous integration setup, and remote sync for synchronizing remote indexes.

Who is it for?

Unch targets developers and engineers who need to perform semantic code searches. Its utility is in finding functions, methods, types, classes, interfaces, and their associated documentation by describing code behavior, rather than using precise names or file paths. This approach helps users discover relevant code even without knowing exact identifiers. Skip this if you do not require semantic code search based on code behavior descriptions.

How to get started

The server authenticates using token-based authentication for external providers. Setup for this model specifically requires using the "unch auth" command to configure tokens for the desired provider.

Use cases

• Local-first code search: Indexing a local repository and performing semantic searches, for example, "unch index --root ." followed by "unch search "create a new router"".
• Remote model-powered search: Authenticating with a provider like OpenRouter and utilizing their embedding models for semantic search, involving commands such as "unch auth openrouter --token sk-or-...", then "unch index --root . --provider openrouter --model openai/text-embedding-3-small", and finally "unch search --provider openrouter --model openai/text-embedding-3-small "create a new router"".
• CI/CD integrated shared indexing: Setting up a CI configuration bound to a GitHub repository, synchronizing remote indexes, and subsequently performing searches against the shared index, using commands like "unch create ci", "unch bind ci https://github.com/owner/repo", "unch remote sync", and "unch search "create a new router"".

Unch offers a specialized solution for semantic code search, suitable for developers and engineers. Its Go foundation and MCP adherence provide a solid framework for integrating with various AI clients. The program supports both local and remote search capabilities, including CI/CD integration for shared indexing.