# Repowise — MCP server card for AI agents

Codebase intelligence for AI coding agents: graph, git history, docs, decisions, code health.

Registry name: `dev.repowise/repowise` · category: Git, GitHub and code review · version 0.61.0 · updated 2026-11-13

## Connect
- PyPI (stdio): `uvx repowise`
- Remote (Claude Code): `claude mcp add --transport http repowise https://api.repowise.dev/mcp`

## Facts
- Repository: https://github.com/repowise-dev/repowise (8,170 stars)
- Website: https://www.repowise.dev

Before you connect a server: read its repository, and give it only the permissions and keys it needs.

## Links
- AgentGid page: https://agentgid.com/mcp/servers/dev-repowise-repowise/


## Repowise: a map of your code for AI assistants

Repowise reads your whole codebase and its change history once, on your own computer, and keeps what it learned in a local index. Your AI coding assistant then asks that index what calls what, what may break and which tests to run, instead of opening file after file. The basic setup is free, needs no key and makes no calls to an AI model.

### Where to use it
- A tour of an unfamiliar codebase: You inherit a project or open someone else's code. Ask your assistant for a tour and it gets the main parts, entry points and riskiest files in one answer.
- Check what a change may break: Before you edit an important file, your assistant asks which other files depend on it, which files usually change with it and which tests to run.
- Review a branch before merging: Ask what your branch made worse and which tests cover the changed lines. The answer comes from the code and its history, with no AI model, so the same change gets the same answer.
- Decide what to clean up first: Get a ranked list of the weakest files, with concrete cleanup plans, and a list of code that nothing uses any more.
- Find out why code is built this way: Before rewriting something odd, ask why it is like that. Repowise looks for recorded decisions and falls back to the change history.
- Shorter sessions on a big project: On a large codebase where the assistant keeps rereading the same files, let it ask the index first. This can save time and usage on long tasks. (idea, not from the official pages)

### How to start
- Install Repowise with uv, pipx or pip. It needs Python 3.11 or newer and Git.

```
uv tool install repowise
```

- In the project folder, build the free index. This run never asks questions and never calls a model.

```
repowise init --yes --no-prose
```

- Restart the assistant (or run /mcp in Claude Code) so it sees the new tools, then ask for get_overview.

### License
Free under the AGPL-3.0 license for people, teams and companies that use it inside their own work. Building it into a product you sell to others without the AGPL duties needs a paid license from Repowise.

### Good to know
- Setup changes files outside your project by default, including Claude settings and a commit hook. Pass --no-editor-setup to keep everything inside the project folder.
- It sends anonymous usage counts (command names, versions, operating system) to Repowise unless you turn them off. No code, file paths or names are sent.
- Written documentation and some extras use an AI model with your own key, which costs money. Code excerpts then go to that model provider.
- The first index of a big project can take several minutes. The local dashboard also needs Node 20 or newer and downloads about 50 MB on first start.
- Most tools only read. Two extra tools, off by default, can write a verdict into the index or call your AI model.

Checked 2026-11-10 against: https://github.com/repowise-dev/repowise/blob/db12feee74bbff347ea66344c23c113517ed21f6/README.md, https://github.com/repowise-dev/repowise/blob/db12feee74bbff347ea66344c23c113517ed21f6/server.json, https://github.com/repowise-dev/repowise/blob/db12feee74bbff347ea66344c23c113517ed21f6/docs/agent/MCP_TOOLS.md, https://github.com/repowise-dev/repowise/blob/db12feee74bbff347ea66344c23c113517ed21f6/docs/agent/INTEGRATIONS.md, https://github.com/repowise-dev/repowise/blob/db12feee74bbff347ea66344c23c113517ed21f6/docs/start/QUICKSTART.md, https://github.com/repowise-dev/repowise/blob/db12feee74bbff347ea66344c23c113517ed21f6/docs/reference/TELEMETRY.md, https://github.com/repowise-dev/repowise/blob/db12feee74bbff347ea66344c23c113517ed21f6/packages/server/src/repowise/server/mcp_server/__init__.py, https://github.com/repowise-dev/repowise/blob/db12feee74bbff347ea66344c23c113517ed21f6/packages/core/src/repowise/core/registry/mcp_tool_registry.py, https://pypi.org/pypi/repowise/json, https://pypi.org/pypi/repowise/0.60.1/json, https://docs.repowise.dev/llms.txt

### Mistakes to avoid
- Don't: Run repowise init on a repo the person only wants to try, which also edits ~/.claude.json, Claude hooks and the git post-commit hook → Do: repowise init --yes --no-prose --no-editor-setup (or REPOWISE_SKIP_EDITOR_SETUP=1) for a trial, CI or sandbox (By default init writes the global Claude entry, hooks and a post-commit hook; the flag keeps writes inside .repowise/.)
- Don't: Call the Repowise tools in the same session right after init → Do: Ask the person to restart the host (or run /mcp in Claude Code), then call get_overview (Hosts read MCP config only at startup.)
- Don't: Run repowise generate (or init without --no-prose while a key is set) without asking → Do: Show repowise generate --dry-run first and run generate only after the person approves the cost (Without --no-prose, --yes uses a model when a key is configured, with the cost pre-approved.)
- Don't: Kill repowise init after a fixed tool timeout on a large repo → Do: Run it in the background and wait for the process to exit, then read the exit code (0 ready, 2 gaps, 130 interrupted) (A few thousand files can take several minutes; a cut run leaves the index incomplete.)
- Don't: Call get_execution_flows or set_finding_status on a default server → Do: Opt in first with mcp.tools in .repowise/config.yaml or repowise mcp --tools "+get_execution_flows", then restart (Seven specialist tools are off by default and clients see the surface fixed at server start.)

### Questions
- Does Repowise change my code? No. The default tools only read the index; one opt-in tool drafts a diff with your own model and another records a verdict in the index.
- Do I need a key or an account? No. The free index needs no key and no account; a model key is only for the optional written docs.
- Which assistants work with it? Claude Code and Codex get every feature; VS Code, Cursor, OpenCode, Hermes, GitHub Copilot CLI and Kiro get the tools plus instructions; any other host can paste a printed config.
- How do I turn off the usage counts? Run repowise telemetry disable, or set DO_NOT_TRACK=1 or REPOWISE_TELEMETRY_DISABLED=1.

---
AgentGid's catalog is curated: every entry was checked against the maker's official pages (registry, repository, license and pricing pages) on the date shown; a license value shows the quote it stands on wherever AgentGid has one. Usage numbers come from AgentGid's own scan of public repositories set up for AI coding agents. Claims by makers are reported, not tested by AgentGid.
