
Repowise MCP Server: Setup for Claude Code, Cursor and Codex
Gives Claude Code, Codex and Cursor a ready-made index of your codebase: what depends on what, what is risky, who owns it and why.
Repowise builds a local index of your repo from its code, git history and docs. The MCP server then lets your assistant ask that index about risk, callers, owners, dead code and the reasons behind decisions. The README says no tool edits your code, and indexing needs no API key.
It runs on your machine and the open-source version is AGPL-3.0. The README says it sends anonymous usage numbers that you can switch off, and sends code to a language model only if you turn on prose generation.
Good for
- Developers on a large codebase who want the agent to stop re-reading files for every task
- Reviewers who want change risk, test gaps and likely owners before they merge
- Teams that want dead code and code health scores from the same index
Works in
12 should work
Tap an app to see how to set it up and any known problems.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Client does not support this transport.
Client does not support this transport.
Install Repowise MCP
Claude Code
- 1
Copy this prompt.
Add the Repowise MCP server to Claude Code. Run this command: claude mcp add repowise -- repowise mcp Then check that it loaded and tell me if I need to sign in or restart.
- 2
Paste it into Claude Code. It shows the change and asks before running it.
Requires uv tool install repowise, repowise init in your repo.
Read-only mode: There are no tools that edit your code. Two opt-in tools write to the index or call your own model, and both are off by default.
Repowise is a Python program, so install it first with uv tool install repowise (or pipx or pip), then run repowise init --yes --no-prose inside your repo to build the index. The generated command for this package would only print help, so the config here starts the server with repowise mcp. For Claude Code, repowise init also registers the server for you, so you may not need the config at all. The dashboard in repowise serve needs Node 20 or newer. Other hosts: repowise agents print-config claude-code prints a server entry you can paste.
Built from the package data we hold. Not yet tested by us.
Tools
Summarizes the architecture of your repo
Returns an architecture summary, module map, entry points and git health. The README says to call it first on an unfamiliar codebase.
Try: “Use Repowise get_overview to summarize this repository.”
Read ·
get_overviewAnswers a question about the code with sources
Searches code and docs, expands through the dependency graph and returns one cited answer with a confidence score.
Try: “How does login work in this repo?”
Read ·
get_answerGives a quick card for files, modules or symbols
Returns a summary, signatures, hotspot flag and governing decisions for several targets at once. You can ask for callers, callees, owners and metrics too.
Try: “Give me context on src/auth.py and src/session.py.”
Read ·
get_contextShows the source of one function or class
Returns one indexed symbol's source with exact line bounds.
Try: “Show me the source of auth.py::verify_token.”
Read ·
get_symbolSearches code and docs by symbol, path or idea
Runs a hybrid search over code and documentation.
Try: “Find where we handle rate limiting.”
Read ·
search_codebaseShows how risky a file or change is
Returns hotspots, dependents, co-change partners, owners, test gaps and bug history. Pass changed files to review a pull request.
Try: “What breaks if I change src/auth.py?”
Read ·
get_riskScores what a commit or diff made worse
Compares a commit, range or uncommitted change on defect risk, maintainability and performance, and lists the tests that touch it.
Try: “How risky is my uncommitted change?”
Read ·
get_change_riskExplains why the code is the way it is
Returns architectural decisions with their evidence, and falls back to git history when no decision is recorded.
Try: “Why do we use a queue between the API and the worker?”
Read ·
get_whyLists code nobody uses
Returns unreachable code by confidence tier.
Try: “What dead code can I safely delete?”
Read ·
get_dead_codeShows code health scores and what to fix first
Returns health scores and findings, coverage, trends, doc drift and refactoring plans.
Try: “What should I fix first in this repo?”
Read ·
get_healthFinds how two files are connected (opt-in)
Finds the shortest dependency path between two files or modules. Turn it on with the tools option.
Try: “How is billing.py connected to auth.py?”
Read ·
get_dependency_pathTraces what runs from an entry point (opt-in)
Shows the top entry points with a call trace from each. Turn it on with the tools option.
Try: “What runs when the CLI starts?”
Read ·
get_execution_flowsDrafts a code diff for a refactoring plan (opt-in)
Turns one refactoring plan into code and a unified diff. It calls your own language model. Turn it on with the tools option.
Try: “Draft the diff for the first refactoring plan.”
Network ·
generate_refactoring_codeRecords a verdict on a refactoring plan (opt-in)
Saves a triage verdict for one plan in the index, such as marking it a false positive. Turn it on with the tools option.
Try: “Mark that refactoring plan as a false positive.”
Write ·
set_finding_statusLists the repos a workspace serves (workspace mode)
In workspace mode, returns each configured repo's alias and paths so you can pass one to the other tools. Workspace mode only; on by default there.
Try: “Which repos does this workspace cover?”
Read ·
list_reposScores coupling across services (opt-in, workspace)
Reports how tightly your services depend on each other, the largest cycle and a 1 to 10 architecture score. Workspace mode only, and turned on with the tools option.
Try: “How coupled are our services?”
Read ·
get_architectureLists services a change could reach (opt-in, workspace)
Walks the cross-repo graph from the service or symbol you name and ranks the downstream services reachable from it. The ranking is a heuristic, not proof of breakage. Workspace mode only, and turned on with the tools option.
Try: “Which services depend on the billing service?”
Read ·
get_blast_radiusFinds dependency-rule breaks (opt-in, workspace)
Lists cross-repo dependencies that break the allow and deny rules you declared, plus circular service dependencies. Workspace mode only, and turned on with the tools option.
Try: “Does our architecture still follow our dependency rules?”
Read ·
get_conformance
Read from the source code
Permissions
low risk · 3 things to watch
Low risk
By default the assistant can only read an index of your repo and cannot edit your code. Keep the two opt-in tools off unless you need them.
What it can do
- Read your data
- Create and change things
- Delete things
- Read-only modeHow to turn it on
Watch out for
Opt-in tools can write or call your model
One opt-in tool writes triage verdicts to the index and another calls your own language model to draft a diff. Leave them off unless you need them. github.com
Usage numbers are sent unless you turn them off
The README says the open-source version sends anonymous CLI telemetry, which you can switch off. github.com
Code stays local unless you enable prose
Indexing needs no key. The README says your LLM provider receives code only if you turn on prose generation. github.com
Security advisories
No published security advisories as of Nov 1, 2026.
Technical details
| Turning on read-only | The 10 default tools only read the index. Two opt-in tools can write to the index or call your own model, and they are off by default. |
|---|---|
| Talks to | Anonymous CLI usage numbers (can be switched off). Your own language model provider, only if you turn on prose generation |
| Sign-in and keys | No account or API key for indexing. A model key is only needed if you turn on generated docs and answers. |
Questions
Is Repowise official?
Repowise is made by the Repowise team and is the product the server belongs to. It is not a server for another company's app.
Do I need an API key?
No for indexing. The README says init runs with no key and no spend. A key is only needed if you turn on generated docs and prose answers with your own model.
How do I connect it to Claude Code?
Run the init command in your repo. The README says it wires Claude Code for the repo and adds a global entry. Then restart Claude Code or run /mcp.
Why does the generated command not work?
The package name alone prints help. The server starts with the mcp subcommand, which the config in the install box uses.
Which tools are on by default?
Ten read-only tools: overview, answer, context, symbol, code search, risk, change risk, why, dead code and health. Seven more are opt-in, and list_repos appears in workspace mode. The README counts 10 default plus 7 opt-in; the published package registers 18 in all.
Is it free?
The open-source version is free under AGPL-3.0. The README also lists a commercial self-hosted option and a hosted service at repowise.dev.
Does it need Node?
Only for the dashboard that repowise serve starts. The README says it needs Node 20 or newer.
Information
updated Nov 2, 2026
- Publisher
- Community
- Package
repowise- Registry
- PyPI
- Transport
- stdio
- Version
- 0.59.0
- License
- AGPL-3.0-or-later
- Last commit
- Oct 31, 2026
- Open issues
- 185
- Listed in
- Official MCP registry
Sources
Docs, issue trackers and discussions this page is built from. Issues, quotes and findings above link to their own source.
Show all 2Hide
- Repowise README · github.com
- MCP tools reference · github.com