Awesome MCP › Software Development

helixml/kodit

⭐ 124 Python repository created 2025-05-01

Kodit indexes Git repositories, splits source files into searchable snippets, and serves them to any MCP-compatible coding assistant so that generated code can be grounded in real examples from the user own projects. It also ingests documents: PDFs, Word files, PowerPoint decks and spreadsheets are rasterised and indexed, allowing search across code and documentation together. Several retrieval strategies are exposed as separate MCP tools so the assistant can choose per query: BM25 keyword search, semantic vector search, regex grep, file listing by glob, reading a file by URI, and visual search over document page images using multimodal embeddings. A second group of tools serves optional LLM-generated enrichments, including architecture documentation, public API documentation, detected database schemas, cookbook style usage examples, commit descriptions, and a wiki with a table of contents and individual pages; these require an enrichment provider to be configured. Alongside MCP, a REST API covers search, repository registration, enrichment and indexing status, with interactive docs served by the running instance. Basic operation needs no external services: a built-in embedding model and SQLite storage ship with the server, which can be started from a Docker image or a pre-built binary. The README documents connection snippets for Claude Code, Cursor, Cline and Kilo Code, and notes that Kodit can also be embedded directly as a Go library, which is how the Helix platform integrates it. One README section suggests wording for a project rule that instructs an assistant to call Kodit search tools before writing code; that text is guidance the repository proposes to its users, not part of the software. Licensed under Apache 2.0 and aimed at teams who want their assistants to retrieve from internal codebases.

https://github.com/helixml/kodit

pythonmcp-servercode-searchragsemantic-searchbm25indexing