Deeprepo
Local RAG engine that indexes codebases into a code graph, embeddings and wiki for AI
- Pricing
- Open source
- Platforms
- CLI, API, self-hosted
- Version
- 1.0.0
- Listing updated
- today
Similar MCP servers
About Deeprepo
Deeprepo is an open-source MCP server for developer tools. It is available on CLI, API, self-hosted, Linux, macOS and Windows.
The README says DeepRepo is a Python library that runs retrieval-augmented generation over a local source tree without a separate vector database or hosted service. Each ingest pass writes a SQLite file holding a code graph of classes, functions, imports and call edges, a set of embeddings with full-text search, and a folder of plain-language markdown pages generated per module.
Queries pass through a router that classifies the question as navigation, impact, explanation, debugging, review or general, then picks a context strategy meant to keep token usage low. The project also ships a CLI with ingest, query, wiki and serve commands, a local wiki viewer with chat, and a set of seven MCP tools so Cursor or Claude Desktop can call it directly.
It suits developers who want to ask questions about a repository with a model of their choice, including fully local models through Ollama, and who would rather keep their code index on their own machine.
Tools this server provides (7)
Taken from the project documentation; the running server may differ.
- ingest_codebase
- One-time setup that indexes a repository directory.
- find_symbol
- Locates a class or function definition.
- get_file_structure
- Shows a file's API without reading the full source.
- explain_file
- Explains what a file does.
- find_change_impact
- Shows what may break if a file changes.
- ask_codebase
- Answers open-ended questions about the code.
- get_project_overview
- Returns an overview of the project.
What it does
- Code knowledge graph
- Stores classes, functions, imports and call edges in SQLite for symbol lookup and blast-radius queries.
- Multi-provider models
- Supports Ollama, OpenAI, Anthropic, Gemini and HuggingFace for both embeddings and answers, configurable separately.
- Seven MCP tools
- Exposes ingestion, symbol search, file structure, explanations, change impact and overview tools to MCP clients.
- Browsable wiki viewer
- Serves generated module pages over HTTP with search and an in-page chat box.
- Branch isolation
- Keeps a separate database per git branch and can seed feature branches from a base branch.
- Incremental indexing
- Skips files whose SHA-256 hash has not changed and reprocesses only the differences.
Good for
- Asking where a symbol is defined in a large unfamiliar repository
- Estimating which files are affected before editing a module
- Letting an AI coding assistant query a codebase without reading raw files
- Generating module-level documentation for an existing project
- Keeping code search local when source cannot leave the machine
Pricing
Open source
Free MIT-licensed library; paid model providers are optional
Deeprepo FAQ
Does it need a vector database?
No. The README says embeddings, full-text index and graph all live in a single SQLite file per branch, with no Chroma, Redis or Postgres required.
Can it run without a paid API?
Yes. The README lists Ollama as a free local option for both the language model and embeddings.
Which MCP clients are mentioned?
The README gives setup examples for Cursor and Claude Desktop and says any MCP-compatible assistant can connect.
What licence is it under?
The README states an MIT License.
Is Deeprepo free?
Yes. Deeprepo is open source, so you can use and self-host it for free.
What are the best alternatives to Deeprepo?
The closest alternatives to Deeprepo are Code-Graph-RAG, Devrag, Codebase Memory, rag-rat and Ai Architect MCP Codebase.
Where this comes from
- Description: written from the project's README and its MCP Registry entry.
