SkillAgentSearch skills...

coderover

Graph-native AI copilot for large codebases. Reviews PRs, answers questions with citations, plugs into Claude Code / Cursor / Aider / Codex / Gemini CLI via MCP.

Install / Use

claude mcp add MUST-Company-AX-Booster -- npx -y github:MUST-Company-AX-Booster/coderover

If the server publishes to npm under a different name, use that package instead — check the repo README.

About this skill
🔌

MCP Server

Model Context Protocol server

Quality Score

83/100

Supported Platforms

Claude Code
Claude Desktop
Cursor
GitHub Copilot
Gemini CLI
OpenAI Codex
Aider

Our assessment of coderover

coderover scores 83/100 on our quality scale, 665th of 949 AI & Machine Learning skills we index.

Its MCP Server is 21 KB long, well organised into 35 sections with 13 code examples: a thorough specification that gives an agent plenty to work with.

It has 3 GitHub stars, so there is little community track record yet; judge it on its content.

Substance
30/30
Structure
20/20
Description
15/15
Adoption
3/20
Freshness
15/15

Maintenance, license and trust

  • The repository was last updated about 3 months ago, so coderover is actively maintained.
  • It is released under the Apache-2.0 license, a permissive license that allows use, modification and commercial use with attribution.
  • Its trust signals score 92/100, with 1 caution from licensing, adoption, age or documentation. These come from repository metadata, not a code audit — read the skill file before letting an agent act on it.

coderover compared with similar skills

All 4 of these similar skills score higher than coderover; compare them before choosing.

SkillScoreStarsUpdatedFormat
coderover (this skill)by MUST-Company-AX-Booster8333mo agoMCP Server
claude-memby thedotmack100101.9ktodayCLAUDE.md
Agent-Reachby Panniantong100100.1ktodayCLAUDE.md
Understand-Anythingby Egonex-AI10087.2k8d agoCLAUDE.md
headroomby headroomlabs-ai10076.5ktodayCLAUDE.md

Frequently asked questions

How do I install coderover?
Run claude mcp add MUST-Company-AX-Booster -- npx -y github:MUST-Company-AX-Booster/coderover. The install tabs above show the steps for each supported agent.
Which AI agents does coderover work with?
It is written for Claude Code, Claude Desktop, Cursor, GitHub Copilot, Gemini CLI, OpenAI Codex and Aider, as a MCP Server file. Other agents that read the same format can often use it too.
Is coderover safe to use?
It is Apache-2.0-licensed and scores 92/100 on trust signals. Skills are instructions an agent will follow, so read the file before installing it and do not approve commands you do not understand.
Is coderover still maintained?
The repository was last updated about 3 months ago, so coderover is actively maintained.
<h1 align="center">CodeRover</h1> <p align="center"> <strong>Graph-native AI copilot for large codebases.</strong> <br /> <em>An autonomous fleet of five agents that review PRs, patrol security, refactor dead code, and answer with cited evidence.</em> <br /> <em>MCP-wired into Claude Code, Cursor, Aider, Codex, and Gemini CLI.</em> </p> <p align="center"> <a href="https://www.star-history.com/#MUST-Company-AX-Booster/coderover"> <picture> <source media="(prefers-color-scheme: dark)" srcset="https://api.star-history.com/badge?repo=MUST-Company-AX-Booster/coderover&theme=dark" /> <source media="(prefers-color-scheme: light)" srcset="https://api.star-history.com/badge?repo=MUST-Company-AX-Booster/coderover" /> <img alt="Star History Rank" src="https://api.star-history.com/badge?repo=MUST-Company-AX-Booster/coderover" /> </picture> </a> </p> <p align="center"> <a href="#-quick-start"><img src="https://img.shields.io/badge/Quick_Start-EDEBE5?labelColor=0A0A0A" alt="Quick Start" /></a> <a href="./LICENSE"><img src="https://img.shields.io/badge/License-Apache_2.0-9FE0B4?labelColor=0A0A0A" alt="License: Apache 2.0" /></a> <a href="https://www.npmjs.com/package/@coderover/mcp"><img src="https://img.shields.io/npm/v/@coderover/mcp.svg?label=%40coderover%2Fmcp&color=EDEBE5&labelColor=0A0A0A" alt="npm: @coderover/mcp" /></a> <a href="https://docs.anthropic.com/en/docs/claude-code"><img src="https://img.shields.io/badge/Claude_Code-D97757?labelColor=0A0A0A" alt="Claude Code" /></a> <a href="#cursor"><img src="https://img.shields.io/badge/Cursor-EDEBE5?labelColor=0A0A0A" alt="Cursor" /></a> <a href="#aider"><img src="https://img.shields.io/badge/Aider-8C8C87?labelColor=0A0A0A" alt="Aider" /></a> <a href="#codex"><img src="https://img.shields.io/badge/Codex-EDEBE5?labelColor=0A0A0A" alt="Codex" /></a> <a href="#gemini-cli"><img src="https://img.shields.io/badge/Gemini_CLI-4285F4?labelColor=0A0A0A" alt="Gemini CLI" /></a> <a href="https://github.com/MUST-Company-AX-Booster/coderover/discussions"><img src="https://img.shields.io/github/discussions/MUST-Company-AX-Booster/coderover?color=8C8C87&labelColor=0A0A0A" alt="Discussions" /></a> <a href="https://coderover.must.company"><img src="https://img.shields.io/badge/Homepage-coderover.must.company-EDEBE5?labelColor=0A0A0A" alt="Homepage" /></a> </p> <p align="center"> <img src="assets/landing-hero.png" alt="CodeRover — graph-aware AI that actually understands your codebase" width="900" /> </p> <p align="center"> <strong>💬 <a href="https://github.com/MUST-Company-AX-Booster/coderover/discussions">Join the discussion &rarr;</a></strong> <br /> <em>Usage questions, design conversations, "how do I…" — all welcome.</em> </p>

[!TIP] [scout] reviewed 218 PRs across 14 design-partner repos this month. If CodeRover saves your team a few hours of context-switching per week, that's the goal. Star the repo if it lands.

You opened a 200,000-line monorepo. Cursor knows the file. Copilot autocompletes the line. Neither of them knows that the function you just edited has 412 callers across six services — and one of them ships PII to the audit log.

CodeRover is an autonomous fleet of five AI agents — [scout], [tinker], [sentinel], [beacon], and [archive] — that operate against a graph of your code, not a window of it. Memgraph stores files, classes, methods, and the edges between them. Every edge carries a confidence tag, so the LLM can cite truth and the graph can flag a hallucination. Plug the fleet into your agent of choice over MCP, or run the dashboard self-hosted on your own infra.

Graph-aware. Self-hosted. Five rovers. One clean room.


✨ Features

[!NOTE] Want to skip the reading? The landing page at coderover.must.company ships a brand-film overview and live screenshots. The /design-system route shows every UI primitive in isolation.

Reason against your code graph

Tree-sitter walks 16 languages and writes the structural skeleton — files, symbols, classes, methods, imports, calls, inheritance — into Memgraph. Every edge is tagged EXTRACTED, INFERRED(score), or AMBIGUOUS, so when an agent answers a question you can tell what came from the AST and what came from a probabilistic guess. Citations on every answer are file:line_start-line_end pairs, not paraphrase.

<p align="center"> <img src="assets/orbital-map.png" alt="Orbital Map — explore the code graph; click any node for citations and graph-walk impact" width="850" /> </p>

Catch regressions before they land

[scout] runs as a GitHub check_run on every PR. It walks the change against the graph, surfaces downstream callers, runs [sentinel] for security findings in the same pass, and either blocks, warns, or approves. Every finding shows agent · severity · file:line · category — never paraphrase, never marketing prose, and never a finding without a path you can click.

<p align="center"> <img src="assets/pr-review.png" alt="PR review — graph-aware findings with file:line citations and downstream impact" width="850" /> </p>

Install into any agent in one command

@coderover/mcp is one npx command. It atomically writes a config block into your agent's MCP config file, mints a scoped JWT against your CodeRover org, and installs seven graph-aware tools. Run remote (HTTP proxy to a self-hosted API) or offline (embedded SQLite + sqlite-vec + tree-sitter + MiniLM, zero network).

<p align="center"> <img src="assets/mcp-install.png" alt="One npx command installs CodeRover into Claude Code, Cursor, Aider, Codex, or Gemini CLI" width="850" /> </p>

Operate the fleet from a brand-locked Mission Control

The dashboard is React + Vite + shadcn/ui under a custom design system — bone on void, BOKEH wordmark, brutalist border-radius 0, brand-film scanlines. Every primitive (Wordmark, Eyebrow, Kicker, Terminal, RoverBadge, AgentStatusLine, CompareTable) is exposed at the public /design-system route as a living spec, so designers and contributors never drift from the tokens in DESIGN.md. The screenshot below is rendered from the live frontend.

<p align="center"> <img src="assets/app-design-system.png" alt="Live screenshot of the /design-system route — wordmark, fleet strip, terminal mockup, status lines, compare table" width="850" /> </p> <table> <tr> <td width="50%" valign="top"> <h3>🛰️ Five-Agent Fleet</h3> <p><code>[scout]</code> reviews PRs · <code>[tinker]</code> proposes refactors · <code>[sentinel]</code> patrols security · <code>[beacon]</code> answers chat with citations · <code>[archive]</code> remembers decisions across PRs.</p> </td> <td width="50%" valign="top"> <h3>🧭 Mission Control Dashboard</h3> <p>React + Vite frontend with Orbital Map (React Flow), terminal-style chat, PR review feed, search with file:line citations, fleet status, and a public <code>/design-system</code> route.</p> </td> </tr> <tr> <td width="50%" valign="top"> <h3>📐 Confidence-Tagged Edges</h3> <p>Every graph edge carries <code>EXTRACTED</code> (AST), <code>INFERRED(score)</code> (LLM), or <code>AMBIGUOUS</code> (no evidence). Confidence glyphs (solid / half / hollow) read in grayscale and pass WCAG AA.</p> </td> <td width="50%" valign="top"> <h3>🔁 Incremental Ingest</h3> <p>Hash-diffed re-ingest — only changed files re-analyze. Rename-aware (preserves edges via qualified-name identity). Watch daemon honors <code>.gitignore</code> with 500ms debounce.</p> </td> </tr> <tr> <td width="50%" valign="top"> <h3>🏢 Multi-Tenant by Default</h3> <p>Per-org scoping on every tenant table. Scope-gated JWTs for MCP. Monthly token caps per org. OpenTelemetry + Prometheus out of the box.</p> </td> <td width="50%" valign="top"> <h3>🧠 Decision Memory</h3> <p><code>[archive]</code> persists architectural decisions and surfaces them in PR review. New PRs that contradict a prior decision get flagged before merge — never re-litigate the same call.</p> </td> </tr> <tr> <td width="50%" valign="top"> <h3>🌐 16 Tree-Sitter Languages</h3> <p>JS / TS, Python, Go, Rust, Java, Kotlin, PHP, Swift, Ruby, C, C++, C#, HTML, CSS, Bash, and more — same graph schema across all of them.</p> </td> <td width="50%" valign="top"> <h3>🪞 Offline Mode</h3> <p><code>@coderover/mcp-offline</code> ships a MiniLM embedder + SQLite-vec + tree-sitter pipeline so the fleet runs against your repo with zero network calls. Air-gapped friendly.</p> </td> </tr> </table>

🚀 Quick Start

1. Install the MCP package into your agent

npx @coderover/mcp install claude-code
# Also supported: cursor, aider, codex, gemini-cli

The installer writes an atomic config update to your agent's MCP config file and walks you through getting a scoped token. Add --offline to run with no backend (uses @coderover/mcp-offline).

2. Or self-host the whole stack

git clone https://github.com/MUST-Company-AX-Booster/coderover.git
cd coderover
cp coderover-api/.env.example coderover-api/.env   # set JWT_SECRET, GITHUB_*, OPENAI_API_KEY
docker compose -f coderover-api/docker-compose.yml up -d

Brings up Postgres 16 + pgvector, Redis 7, Memgraph, the NestJS API, and the React frontend. Full first-run, env vars, and migrations are in SETUP.md.

3. Ingest a repo

# From the dashboard at http://localhost:5173/repos — click "Add repository"
# or via the MCP surface inside your agent:
rover ingest /path/to/your/repo

A multi-agent pipeline scans the repo, extracts every file, function, class, and dependency, and writes the graph to Memgraph + embeddings to pgvector. Incremental — re-runs only re-analyze what changed.

4. Open Mission Control

open http://localhost:5173

The dashboard shows the fleet ([scout] [tinker] [sentinel] [beacon] [archive]), the Orbital Map of your code graph, the PR review feed, search with file:line citations, and the live /design-system spec.

# In your agent, ask anything against the graph:
> what calls authenticate()?
> diff impact for branch fix/oauth-state
> who decided we use refresh tokens in cookies?
> /search anomaly detection in the watch daemon

🌐 Multi-Platform Installation

CodeRover speaks MCP. Any agent that supports MCP can call the seven graph-aware tools.

Claude Code

npx @coderover/mcp install claude-code

Writes to ~/.claude/claude_desktop_config.json. Restart Claude Code to load. The tool surface appears under coderover in the MCP picker.

Cursor

npx @coderover/mcp install cursor

Writes to ~/Library/Application Support/Cursor/User/globalStorage/mcp.json (mac) or the equivalent on Linux/Windows. Reload the Cursor window.

Aider

npx @coderover/mcp install aider

Adds the CodeRover MCP server to ~/.aider.conf.yml. Run aider --mcp to load.

Codex

npx @coderover/mcp install codex

Adds the server to ~/.codex/mcp.json. Codex picks it up on next launch.

Gemini CLI

npx @coderover/mcp install gemini-cli

Writes to ~/.gemini/mcp_servers.json. Tools land under coderover.*.

Offline mode (any client)

npx @coderover/mcp install <agent> --offline
# or pin the offline embedder package directly:
npm i -g @coderover/mcp-offline

Embedded SQLite + sqlite-vec + tree-sitter + MiniLM. Zero network calls after install. Useful for air-gapped envs and local development.

VS Code extension

The CodeRover VS Code extension (in coderover-api/vscode-extension/) ships chat + PR review inside the editor with SSE streaming and SecretStorage-backed tokens. Install from the Marketplace or sideloa

Truncated for display — read the full file on GitHub.

Related Skills

View on GitHub
GitHub Stars3
CategoryAI
Updated2mo ago
Forks0

Languages

TypeScript

Trust signals

92/100

From repository metadata: license, adoption, age and documentation. Not a code audit — see the Safety scan above for what the skill file itself contains.

1 low