</>MCP Agents Market
MCP Server

repowise

by repowise-dev6.3kPythonUpdated 2026-08-28

Codebase intelligence for AI and humans: code health scores, auto-generated docs, git analytics, dead code detection, and architectural decisions via MCP.

Claude CodeCodexCursorVS CodeOpenCodeHermes

Repowise is an MCP server that delivers comprehensive codebase intelligence to AI coding agents through ten structured tools. It indexes code dependency graphs, git history, tests, documentation, and architectural decisions once, then serves cited answers, change impact analysis, and code-health diagnostics to agents via MCP. The engine runs locally with zero LLM calls for graph, risk, health, and PR review layers, while optional model-written documentation can upgrade the structural wiki. Agents using Repowise cut output by 31.6% and reach answers in 3.8 tool calls instead of 7.2 by receiving task-shaped context rather than rediscovering repository structure on every question.

Key Features

Ten MCP tools exposing graph, git, docs, decisions, and health intelligence in task-shaped payloads
AST-parsed dependency graphs across 26 languages with symbol-level resolution and confidence scoring
Zero-LLM deterministic analysis for graph, git, code health (49 detectors), change risk, and dead code detection
Command-output distillation that compresses shell output 61-89% before agents read it, reversibly
Code health scoring (1-10) with concrete refactoring plans (Extract Class, Break Cycle, etc.) ranked by blast radius
Multi-repository workspace intelligence: cross-repo contracts, breaking-change guards, federated MCP endpoint
GitHub PR bot posting symbol-level blast radius, change risk, missing tests, and co-change partners with zero LLM calls
Local dashboard and VS Code extension serving the same index alongside MCP tools for agents

Use Cases

  • 01Provide coding agents (Claude Code, Codex, Cursor) with graph-grounded context so they find the right files in fewer calls
  • 02Assess pull request risk with symbol-level impact, missing co-change partners, and test coverage before merge
  • 03Identify defect-prone files and receive dataflow-verified refactoring plans with exact methods and blast radius
  • 04Detect dead code, untested hotspots, and cross-file performance risks (N+1, I/O-in-loop) via call-graph analysis
  • 05Understand cross-repository breaking changes by matching backend contracts to frontend consumers in workspaces
  • 06Auto-generate and maintain repository documentation (wiki, CLAUDE.md, AGENTS.md) from structural analysis or optional LLM prose

Related MCP Servers

View more

repowise — FAQ

What is Repowise?+

Repowise is a self-hosted MCP server that indexes your codebase's dependency graph, git history, tests, documentation, and architectural decisions, then exposes ten task-shaped tools to AI coding agents over the Model Context Protocol. Core analysis layers (graph, git, health, change risk) require zero LLM calls; optional documentation prose can use your provider or run offline via Ollama.

How do I install and connect Repowise to my agent?+

Install via pip (`pip install repowise`), run `repowise init` in your repository to build the index, then `repowise serve` starts both the MCP server and local dashboard. For Claude Code, `init` auto-wires the config; for other MCP clients (Codex, Cursor, VS Code) add a JSON entry pointing to `repowise mcp` as the command. Full wiring for six agents is documented in the quickstart.

Which AI clients and agents does Repowise support?+

Repowise integrates at 'Full' tier (MCP tools, hooks, transcript mining) with Claude Code and Codex CLI, and 'Good' tier (MCP tools only) with VS Code, Cursor, OpenCode, and Hermes. Any other MCP-compatible client can connect via a single config snippet. The VS Code extension also exposes six tools to GitHub Copilot.

Does Repowise require API keys or send code to the cloud?+

The core index (graph, git, health, change risk, dead code, PR review) runs locally with zero LLM calls and no API key. Optional model-written documentation needs a provider key (Anthropic, OpenAI, Gemini) or can run fully offline via Ollama. In keyless mode (`repowise init --no-prose -y`) nothing leaves your machine and the wiki is rendered from code structure.

Is Repowise free, and what languages does it support?+

Repowise is open source (AGPL-3.0) and free for internal use by individuals, teams, and companies; commercial licensing covers enterprise features (SSO, RBAC, SLA support, embedding in products). It parses 26 languages to AST (Python, TypeScript, Java, Go, Rust, C++, C#, etc.) and tracks 40 total across five support tiers; all languages ship in the open-source distribution.

What are the ten MCP tools Repowise provides?+

The tools are: get_overview (architecture summary), get_answer (hybrid retrieval with citations), get_context (file/symbol triage cards), get_symbol (exact source spans), search_codebase (semantic wiki search), get_risk (hotspots, dependents, co-change), get_change_risk (commit/PR scoring), get_why (architectural decisions), get_dead_code (unreachable code), and get_health (code health scores and refactoring plans). Each accepts batch targets and returns task-shaped payloads with metadata on index freshness.

How do I install repowise?+

Open the source repository on GitHub and follow its README. repowise is a mcp server — MCP Agents Market links you directly to the official repo.

Is repowise free?+

repowise is an open-source project hosted on GitHub. Check the repository for its license and any usage requirements.

Related searches