Repowise

Repowise is a codebase intelligence tool from repowise-dev. It indexes a repository's structure, git history, tests and architectural decisions into one index, then serves that index to coding agents over MCP, to developers through a dashboard and CLI, and to reviewers through a pull request bot. The core engine is open source under AGPL-3.0, with a hosted service and a self-hosted commercial edition for organizations.

What It Is

Repowise builds a call graph and dependency graph across 28 AST-parsed languages and combines it with git analytics, ownership, test reach and code-health scoring. The README states that graph, git, health, change risk, test and dead-code analysis make zero LLM calls, so indexing needs no API key; generated prose docs and answers are optional and use a model provider of your choice, including local models through Ollama.

Typical workflow: install the CLI (uv tool install repowise or pip/pipx), run repowise init in a repository, then repowise serve to start the dashboard and MCP server. Agents such as Claude Code, Codex and Cursor connect to the MCP server, and repowise init wires up Claude Code for the repo.

How It Serves Agents

The MCP server exposes ten task-shaped tools, including get_overview, get_answer, get_context, get_symbol, search_codebase, get_risk, get_change_risk, get_why, get_dead_code and get_health. Each response carries the indexed commit and a stale-index warning. Optional hooks, a generated CLAUDE.md/AGENTS.md, and repowise distill for trimming command output are also documented. The project reports a benchmark of 31.6% less agent output on django/django (n=43) as its own measurement.

Change Safety and Code Health

For a diff or pull request, Repowise reports the callers a change reaches, companion files that usually change with it but are missing, and the tests that reach it. The PR bot posts one comment per pull request and stays silent on clean ones. Each merged change can become a Change Record with owners, tests, decisions crossed and policy checks run in CI via repowise check.

Every file receives a 1 to 10 health score from deterministic detectors covering defect risk, maintainability and performance, with ranked fix-first queues, refactoring plans and dead-code detection. Further capabilities include decision mining from ADRs, commits and PRs, doc-drift checks, secret scanning across git history, test impact without a coverage run, and multi-repository workspaces with cross-repo contract mapping.

Deployment Options

The open-source package runs on your machine. The hosted service at repowise.dev re-indexes on push, adds the PR bot, private repositories and MCP access for Claude.ai and ChatGPT. A commercial edition runs in a customer VPC or air-gapped. A VS Code extension and a Claude Code Lens pane are also documented. Foresight, a model-based predictor of files a fix will touch, is listed as coming to hosted and enterprise.