# Mracos Mcp

> smol utility for using MCP tools across many coding agents

- **Type:** MCP server
- **Install:** `agentstack add mcp-mracos-mcp`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [mracos](https://agentstack.voostack.com/s/mracos)
- **Installs:** 0
- **Category:** [Integrations](https://agentstack.voostack.com/c/integrations)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [mracos](https://github.com/mracos)
- **Source:** https://github.com/mracos/mcp

## Install

```sh
agentstack add mcp-mracos-mcp
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# mcp

Manage MCP servers from a single config file (`~/.mcp-servers.json`) and sync to multiple AI coding clients.

Supported targets:
- **Claude Code** - `~/.claude.json` (`mcpServers`)
- **Codex** - `~/.codex/config.toml` (`[mcp_servers.*]`)

## Install

**zinit**

```sh
# as command (in a `zinit for as"command"` block)
pick"bin/*" mracos/mcp

# or standalone
zinit ice as"command" pick"bin/*"
zinit light mracos/mcp
```

**Clone + PATH**

```sh
git clone https://github.com/mracos/mcp ~/.mcp
export PATH="$HOME/.mcp/bin:$PATH"
```

## Requirements

- **bash** 3.2+ (macOS stock bash works)
- **[jq](https://jqlang.github.io/jq/)** - JSON processing (all commands)
- **Node.js / npx** - daemon management (`mcp daemon`, `stdio-http-proxy` servers)
- **[mcp-proxy](https://github.com/nicholasgasior/mcp-proxy)** - stdio-to-HTTP proxy (`npx mcp-proxy`)
- **envsubst** (gettext) - env var expansion in daemon configs
- **[launcher](https://github.com/mracos/launcher)** - only if you set `MCP_BACKEND=launchd`

## Usage

```bash
mcp list                          # List all configured servers
mcp add notion https://...        # Add HTTP server
mcp add tool -- npx -y @pkg/mcp   # Add stdio server
mcp remove                  # Remove a server
mcp edit                          # Open in $EDITOR
mcp apply                         # Sync into clients
mcp show                    # Show server config
mcp daemon [start|stop|status]    # Manage proxy daemons
mcp logs [name]                   # Tail daemon logs
```

## Server Types

### HTTP

Direct connection to a remote MCP server:

```bash
mcp add notion https://mcp.notion.com/mcp
```

### stdio

Spawns a new process per session:

```bash
mcp add mytool -- npx -y @example/mcp
```

### stdio-http-proxy

stdio servers spawn a new process for every Claude Code session - slow startup, no shared state, duplicate processes if you have multiple windows. `stdio-http-proxy` runs the server once as a persistent daemon via [mcp-proxy](https://github.com/nicholasgasior/mcp-proxy) and connects clients over HTTP:

```json
{
  "myserver": {
    "type": "stdio-http-proxy",
    "command": "npx",
    "args": ["-y", "@example/mcp"],
    "env": { "API_KEY": "${MY_API_KEY}" },
    "port": 8081
  }
```

Use unique ports per server. `mcp apply` starts daemons automatically and writes SSE endpoints to Claude Code config.

### Why SSE instead of HTTP?

Claude Code assumes `http` endpoints support OAuth and tries discovery. Local `mcp-proxy` doesn't implement OAuth, causing 404 errors. `sse` transport skips OAuth discovery and connects directly.

For Codex, proxied servers are written as streamable HTTP URLs (`http://localhost:/mcp`).

## Env Vars

Add env vars with `--env` on `mcp add`:

```bash
mcp add --env API_KEY=sk-123 mytool -- npx -y @example/mcp
```

Or edit directly with `mcp edit`:

```json
{
  "myserver": {
    "type": "stdio",
    "command": "npx",
    "args": ["-y", "@example/mcp"],
    "env": { "API_KEY": "${MY_API_KEY}" }
  }
}
```

`${VAR}` references in env values are expanded via `envsubst` when generating daemon configs.

## How It Works

`~/.mcp-servers.json` is the source of truth. `mcp apply` reads it and:

1. Starts daemons for `stdio-http-proxy` servers (via the active backend)
2. Merges server configs into `~/.claude.json`
3. Syncs server configs into `~/.codex/config.toml`

Existing non-MCP keys in both target files are preserved.

### Daemon backends

`mcp daemon` dispatches to a pluggable backend selected by `MCP_BACKEND`:

| Backend | When to use | Notes |
|---|---|---|
| `pm2` (default) | Cross-platform, fast iteration | Requires `npx`; pm2 installed lazily on first use |
| `launchd` | macOS native, survives reboots | Requires the [`launcher`](https://github.com/mracos/launcher) CLI in PATH |

```bash
export MCP_BACKEND=launchd   # or pm2 (default)
mcp daemon start
```

Both backends are *self-contained*: top-level binaries (`npx`, the inner `command`) are resolved to absolute paths at generation time and `PATH` is injected into the daemon's environment so child spawns work even when the launchd/pm2 process starts from a minimal env.

The backend interface is defined in `lib/lib-backend.bash`; see [docs/adrs/0001-mcp-backend-abstraction.md](docs/adrs/0001-mcp-backend-abstraction.md) for the design.

## Testing

```bash
npm install
npm test
```

## Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [mracos](https://github.com/mracos)
- **Source:** [mracos/mcp](https://github.com/mracos/mcp)
- **License:** MIT

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/mcp-mracos-mcp
- Seller: https://agentstack.voostack.com/s/mracos
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.