Repowise MCP
A local codebase index with ten MCP tools: dependency graph, git history, code health and architectural decisions without re-grepping on every step
- 8,126 ★ on the repository
- Last commit
- AGPL-3.0
Extended access
Why this level
- Reads the entire source code and git history of the repository to build the index
- By default can make one model call when capturing decisions from session transcripts
Install
- Works in
Manual install
pip install repowise
cd /path/to/your/repo
repowise init --no-prose -y
repowise serveBuilds the index locally and starts the MCP server, init wires up Claude Code automatically.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Reads the entire source code and git history of the repository to build the index
- By default can make one model call when capturing decisions from session transcripts
This is third-party code. Review the repository files before installing.
What it does
Repowise builds a local repository index once: a file and symbol dependency graph across 26 languages, git history with hotspots and co-change data, a per-module wiki, architectural decisions mined from commit history, and 51 deterministic code-quality detectors. On top of that index it exposes ten MCP tools that hand the agent a ready answer instead of a chain of grep and read calls: who calls a function, what breaks if it changes, which tests to run for a given diff. The repowise distill command compresses the output of commands like pytest or git log, keeping the errors and cutting the noise, with a way to expand the original output back.
Who it is for. For developers who want the agent to already understand the repository's structure and history instead of rebuilding context on every task.
Good fit when
- The agent burns many calls figuring out who calls a function or what breaks when a file changes
- You need to quickly know which tests a given diff actually touches
- Command output like pytest or git log fills the context with noise
Not a fit when
- The repository is small enough that the agent already navigates it quickly
- You need a managed multi-tenant cloud service rather than a local self-hosted index
Example request
Use Repowise get_overview to summarize this repository, and show me what breaks if I change src/auth.pyMCP
- Transport
- stdio
- Authentication
- not required
Limitations
The core analysis needs no model key, but capturing decisions from session transcripts makes one batched model call by default, turned off with --no-llm. Multi-repository workspaces need extra setup. The project is AGPL-3.0, a separate license is needed for commercial use without disclosing source.
How to disable. Remove the package with pip uninstall repowise and stop the local server, the MCP connection comes out with claude mcp remove repowise.
FAQ
Do I need a model key to run it?
No, the graph, git analysis, code health, tests and dead-code detection run without calling a model, generated prose is optional.
Where do session transcripts go?
Nowhere off the machine, processing happens locally, and --no-llm removes the one model call in that process entirely.
Discussion
No questions or reviews yet.
Related
MCP serversLocal code structure graph: builds a repository map with tree-sitter and serves the assistant only the context a change actually touches over MCP
Gives the agent an understanding of your repository: it indexes the code and answers questions about it in Claude Code, Cursor, Codex, Windsurf and more
An MCP server that builds a code graph of functions, classes, calls and routes and answers the agent's structural queries locally
A local codebase index for agents: hybrid search, dependency graphs and change impact analysis
An MCP server builds a graph of code entities, relations and flows so the agent finds the right files without reading everything
Indexes a codebase into a knowledge graph and gives the agent call chains, change blast radius and code search over MCP
README in short
The README presents Repowise as an index of code, git history, tests, docs and decisions built once and then queried by agents and developers with cited sources. It cites measurements on django/django: fewer tool calls and fewer tokens for the same question compared to running without the index. It covers ten MCP tools, generating CLAUDE.md and AGENTS.md from the index, hooks that push context in at the right moment, and the distill command for compressing noisy command output. Install is via pip, self-hosted, AGPL-3.0 or commercial license.