# repowise-dev/repowise [Health: Active]

**Category:** 🔄 Version Control  
**Repository:** https://github.com/repowise-dev/repowise  
**GitHub Stars:** 7928  
**Views:** 6  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/repowise-dev-repowise

## Description
Codebase intelligence for AI coding agents. Indexes a repo into a dependency graph, git history, defect-risk code health, auto-generated docs, and architectural decisions, served as 9 task-shaped MCP tools across 15 languages. Runs 100% local; hosted option available.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `uvx` (confidence: high):

```json
"mcpServers": {
  "repowise": {
    "command": "uvx",
    "args": ["repowise"]
  }
}
```

## Documentation

## What repowise-dev/repowise does

repowise-dev/repowise builds an index of a repository and makes that index available to coding agents through MCP. The indexed material includes source code, Git history, tests, and architectural decisions. It supports codebase analysis across 28 languages and can also produce a generated wiki covering modules and files.

The resulting context is designed for questions that require relationships between files rather than the contents of one file. An agent can investigate what calls a changed symbol, which tests reach it, which related files may be missing from a change, and which areas have the highest defect or maintenance risk. The dashboard presents the same indexed information with architecture maps, C4 views, health scores, charts, and cited answers.

## How it works

The normal flow is to install the Python package, initialize the repository, and start the service. Initialization creates the graph and analysis data locally without an API key. It can include Git information, code health, dead-code analysis, and documentation generation. Model-written documentation can be generated separately when needed.

`repowise serve` starts the dashboard and MCP server. An MCP-capable coding agent then queries the index instead of repeatedly grepping through the repository and opening files. The tool set is organized around repository tasks, including overview, dependency and caller analysis, change risk, tests, health, and architectural context.

Repowise also includes `repowise distill`, which reduces output from tests, installs, and Git commands while retaining error lines. Optional hooks can warn about fragile files and stale paths, and generated `CLAUDE.md` or `AGENTS.md` files can be kept aligned with the index.

## Setup and configuration

Install the package with one of the supported Python tooling options:

```bash
uv tool install repowise
# or: pipx install repowise
# or: pip install repowise
```

From the repository to analyze, initialize the index and start the service:

```bash
repowise init --yes --no-prose
repowise serve
```

The `--no-prose` option initializes the graph and analysis layers without model-written prose. Run `repowise generate` later to add that documentation. The first `serve` run downloads the dashboard interface and requires Node.js 20 or newer. Initialization itself requires no API key. The setup flow can configure Claude Code for the repository; the service is also intended for other MCP-capable agents.

## Tools and capabilities

- Repository overview and indexed codebase questions
- Dependency, caller, and architecture analysis
- Change-risk analysis, including related files and affected tests
- Per-file code-health scoring and prioritized remediation queues
- Git-history analysis tied to defects and code risk
- Dead-code detection and refactoring guidance
- Generated repository and module documentation
- Cross-repository service maps and contract impact checks
- Dashboard answers with citations and charts
- Command-output reduction through `repowise distill`

_Full upstream README: https://allmcps.com/mcp/repowise-dev-repowise/readme_