Omni Server

One MCP server instead of six. GitHub, local Git, Slack, web fetching, persistent memory, and filesystem tools — 15 tools in a single config entry, with dangerous capabilities off by default.

npm version npm downloads License: MIT Model Context Protocol

Why Omni?

Setting up MCP usually means installing and configuring a separate server for every capability — one for GitHub, one for filesystem, one for memory, one for fetch. Each has its own config block, its own runtime, its own startup cost.

Omni bundles the everyday developer tools into one server, one config entry, one npx command:

• 🔒 Safe by default — shell execution and file writes are disabled unless you explicitly opt in to Agent Mode.
• 🏠 Local & private — runs on your machine over stdio. You bring your own tokens; nothing is proxied through a third party.
• 🪶 Zero install — runs straight from npm via npx.

🚀 Quick Start

Interactive setup (Claude Desktop)

npx -y @bitofacoder/omni-mcp-server@latest setup

The wizard asks for your optional tokens (GitHub, Slack), asks whether to enable Agent Mode, and updates your Claude Desktop config for you. Restart Claude Desktop and you're done.

Claude Code

claude mcp add omni -e GITHUB_PERSONAL_ACCESS_TOKEN=your_gh_token -- npx -y @bitofacoder/omni-mcp-server@latest

Manual config (Claude Desktop, Cursor, Windsurf, Codex CLI…)

Add this to your client's MCP config (e.g. claude_desktop_config.json):

{
  "mcpServers": {
    "omni": {
      "command": "npx",
      "args": ["-y", "@bitofacoder/omni-mcp-server@latest"],
      "env": {
        "GITHUB_PERSONAL_ACCESS_TOKEN": "your_gh_token",
        "SLACK_BOT_TOKEN": "your_slack_token"
      }
    }
  }
}

All env vars are optional — tools that need a missing token simply tell you so.

🛠️ Tools (15)

🔒 Agent Mode (opt-in)

system_execute_command and system_write_file let the AI act on your machine — run tests, install packages, edit files. Because that's real power, they are disabled by default and don't even appear in the tool list until you enable them:

"env": {
  "OMNI_AGENT_MODE": "true"
}

Everything else (reading files, git status, web fetch, memory) is read-only or sandboxed to its own data file.

🤝 Contributing

PRs welcome — especially new integrations (Linear, Notion, Jira, Discord) and Agent Mode hardening (allowlists, working-dir scoping). Open an issue first for bigger changes.

If Omni saved you some config wrangling, a ⭐ helps other people find it.

📄 License

MIT — see LICENSE.