Repowise
Give Claude Code or Cursor structured repo context—dependency graph, git history, docs, ADRs, and health signals—without re-scanning the tree every prompt.
Overview
Repowise is a MCP server for the Build phase that exposes indexed repository graph, git history, docs, decisions, and code health to coding agents over stdio.
What is this MCP server?
- Serves codebase graph, git history, docs, decisions, and code-health signals over MCP stdio (PyPI repowise 0.17.1).
- Requires a local index first: run repowise init on the repo, then point the server at that absolute repo path.
- stdio transport via repowise mcp subcommand with --transport stdio for Claude Code, Cursor, and other MCP clients.
- Self-hosted intelligence on your machine—no cloud API key for the local package path.
- Complements raw file reads with structured relationships agents use for safer refactors.
- Package version 0.17.1 on PyPI with stdio MCP transport
- CLI entry: repowise mcp plus required repo_path filepath argument
- Publisher schema: Model Context Protocol server registry 2025-12-11
Community signal: 2.3k GitHub stars.
What problem does it solve?
Agents lose thread on large repos because every turn re-reads files and misses architecture, history, and written decisions.
Who is it for?
Indie developers who keep the repo on disk, want privacy-friendly context, and already use MCP with Claude Code or Cursor.
Skip if: Teams that refuse local indexing, need instant context on a laptop clone without init, or only want hosted multi-repo dashboards.
What do I get? / Deliverables
After indexing and registering MCP, your agent queries a stable intelligence layer for relationships and repo knowledge while you build and change code.
- MCP tools backed by indexed graph, history, docs, decisions, and health data
- Agent-accessible structured repo context during coding sessions
- Repeatable local setup without a repowise.dev API key
Recommended MCP Servers
Journey fit
Canonical shelf is Build because the server is wired into the agent loop while you implement and refactor; it is not a one-off research or launch tactic. agent-tooling fits MCP stdio serving an indexed workspace—the same moment you configure Claude/Cursor MCP and need durable codebase memory.
How it compares
MCP codebase-intelligence server with a required local index—not a hosted-only API and not a generic filesystem MCP.
Common Questions / FAQ
Who is dev.repowise/repowise for?
Solo and small-team builders using AI coding agents who want graph, history, docs, and decision context from a repo they control locally.
When should I use dev.repowise/repowise?
Use it once repowise init has indexed the project and you are wiring MCP during active development, refactors, or review-oriented agent sessions.
How do I add dev.repowise/repowise to my agent?
Install repowise from PyPI (0.17.1), run repowise init on the repo, then add an MCP server entry that runs repowise mcp <absolute-repo-path> --transport stdio.