# RepoRelay [Health: Active]

**Category:** 🔄 Version Control  
**Repository:** https://github.com/Lukie-81/RepoRelay  
**GitHub Stars:** 26  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/reporelay

## Description
Bounded local MCP access to one approved repository for AI read/search and structured handoffs.

## 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": {
  "reporelay": {
    "url": "http://127.0.0.1:7676/mcp"
  }
}
```

## Documentation

## What RepoRelay MCP server does

RepoRelay MCP server creates a local boundary around one repository selected by path. ChatGPT can inspect files, list directory contents, read file contents, and search the repository. The selected directory must be an existing project folder; the server does not expose a drive root, a whole user folder, or unrelated paths.

The default handoff mode adds three writers for predetermined files: one for the next task, one for a review, and one for handoff state. These files support a workflow in which ChatGPT reviews or plans work and a separate local coding agent, such as Codex or Claude, implements it. The writers do not provide general source editing or arbitrary file creation.

## How it works

RepoRelay MCP server runs locally and acts as both the MCP server and the access boundary. ChatGPT selects the available tools, while an OpenAI Secure MCP Tunnel carries traffic to the computer. The local `tunnel-client` forwards that traffic to RepoRelay; it is a network forwarder rather than the component that defines repository permissions.

The normal setup exposes seven tools: four for reading and searching, and three for fixed handoff files. Read-only mode exposes only the four read/search tools. The read tools include `open_workspace`, `list_files`, `read_file`, and `search_files`; `list_files` can also be used to find instruction files with its `instructionsOnly` option.

RepoRelay MCP server does not run shell or PowerShell commands, Git, or other processes. It cannot launch programs, delete files, edit arbitrary source files, choose arbitrary write targets, or inspect locations outside the approved repository.

## Setup and configuration

Install the npm package globally with Node.js version 22.19 or newer and below 27:

```bash
npm install -g reporelay-mcp@latest
```

Start the default handoff mode by passing a quoted repository path:

```bash
reporelay quickstart "$HOME/Projects/my-app"
```

For read-only access, add `--no-handoff-writes`:

```bash
reporelay quickstart "$HOME/Projects/my-app" --no-handoff-writes
```

The quickstart process configures the local service and walks through Secure MCP Tunnel setup. It automatically installs OpenAI's official `tunnel-client`. You also need access to the Secure MCP Tunnel, permission to create a custom MCP app in the target ChatGPT workspace, and ChatGPT developer mode where required by that workspace. The tunnel is configured on the OpenAI platform; the RepoRelay MCP server connection is added inside ChatGPT.

## Tools and capabilities

Available read and search capabilities are:

- `open_workspace` for opening the approved workspace.
- `list_files` for listing files, including an instruction-only view.
- `read_file` for reading repository files.
- `search_files` for finding content in the repository.

The default mode also provides `write_next_task`, `write_review`, and `update_handoff_state`. These write only to the defined handoff destinations. Read-only mode removes all three writers rather than changing the repository boundary.

## Limitations and notes

RepoRelay MCP server depends on OpenAI Secure MCP Tunnel availability, account permissions, and ChatGPT workspace settings. Personal, Business, Enterprise, and Edu workspaces may expose different capabilities. The server cannot enable developer mode or tunnel access for you.

The repository path should always be quoted, especially on Windows. RepoRelay is designed for an existing local project folder, not a machine-wide filesystem view. Its security model limits what ChatGPT can request, but implementation work still belongs to a separate coding agent or developer.

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