# Arroway [Health: Active]

**Category:** 💻 Developer Tools  
**Repository:** https://github.com/oaleviola/arroway-plugin  
**GitHub Stars:** 2  
**Views:** 1  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/arroway

## Description
Arroway orchestrates people and AI agents around the decisions in force.

## Claude Desktop Quick Installation
Remote MCP endpoint (confidence: high). Install path detected from listing signals. Add as a URL/SSE server in your client:

```json
"mcpServers": {
  "arroway": {
    "url": "https://www.arroway.app/api/mcp"
  }
}
```

## Documentation

## What the Arroway MCP server does

Arroway MCP server gives connected assistants access to a shared record of team decisions, project context, norms, activity, and unfinished work. The accompanying workflow skill instructs an assistant to consult that record before changing something and to leave a durable record after completing or pausing a task.

The system separates assistant proposals from decisions made by people. An assistant can record a proposal, but a person must approve it before it becomes a team decision. Personal standing rules can also follow a user across projects without being visible to other people.

## How it works

The workflow centers on four points in an agent task:

- Before acting, the assistant uses `arroway_read` to inspect relevant project decisions and context.
- Before making a claim or recommendation, `arroway_norms` provides the applicable established rules at lower cost than a full read.
- After finishing, `arroway_log` records information that could prevent someone else from repeating the work.
- When stopping early, `arroway_pass` records the current state, one next step, work that should not be repeated, and the remaining risk.

A later assistant can find the handoff during its first read and take it with `arroway_claim`. Direct requests are also supported: `arroway_catch_up` summarizes activity across projects, `arroway_search_log` finds previous work, and `arroway_remember` records a decision stated by the user.

## Setup and configuration

Claude Code users can add the marketplace and install the plugin with:

```bash
claude plugin marketplace add oaleviola/arroway-plugin
claude plugin install arroway@arroway
```

After installation, sign in through Claude Code's `/mcp` interface by selecting `plugin:arroway:arroway` and completing the browser flow. The plugin includes the connection address, so no separate endpoint needs to be pasted. Until sign-in is complete, Arroway tools do not appear and the assistant does not read or write Arroway data.

Cursor users add the repository as a plugin marketplace and install `arroway` from the plugin list. The Cursor package also includes the connection and starts browser-based sign-in when needed.

For agents that can read skills but do not have a dedicated plugin, install the workflow skill with `npx skills add oaleviola/arroway-plugin`. The agent still needs an MCP connection to `https://www.arroway.app/api/mcp`.

## Tools and capabilities

The Arroway MCP server exposes tools for:

- Reading project context and current decisions.
- Retrieving established norms before claims or actions.
- Logging durable results from completed tasks.
- Passing unfinished work to a later agent.
- Claiming a handoff.
- Catching up across projects.
- Searching the activity log.
- Remembering user-approved decisions.

Hooks reinforce the read-before-action and record-after-task workflow. They also provide a check of text before publication and, for Claude Code and Codex, allow a user to answer an agent permission request from the Arroway cockpit. A cockpit approval applies only to the matching, single-use request and does not approve requests automatically.

## Limitations and notes

The repository is a generated mirror; edits made there are overwritten by later publishes. It is intended for installation and reference rather than direct product development.

The plugin requires a browser sign-in. If the Arroway connection is unavailable, the cockpit is off, or a permission response does not match the request fingerprint and session credential, the normal permission prompt remains in place. The browser and phone usage described by the project uses Arroway as a connector rather than this repository's plugin.

_Full upstream README: https://allmcps.com/mcp/arroway/readme_