Install first. Connect your AI after.

1. Install Autonomous

On the computer you want the AI to use, install the public npm package.

npm install -g autonomousmcp@latest && autonomous

The published package is autonomousmcp on npm. The user-facing command is autonomous.

2. Open a new terminal and start it

After npm finishes, open a new terminal window so the global command is available, then run:

autonomous

Keep this terminal running while you want the computer available, unless you explicitly enable autostart.

3. Pairing is automatic

The first run creates a short-lived secure pairing session, copies the pairing link when possible, and opens your browser automatically.

Sign in or create your Autonomous account in that browser flow. After authentication, the browser approves the pending computer and the local client finishes pairing automatically. You do not need to copy a pairing code.

If the browser cannot open, the terminal keeps the secure link visible and copyable so you can open it manually.

4. Connect ChatGPT manually

Until the official Autonomous plugin is available, add the remote MCP server manually in ChatGPT web.

https://mcp.octamundi.com/v2

Open Settings → Apps → Advanced settings and enable Developer mode if your account exposes it. Then, in Settings → Apps, click Create app (Criar aplicativo MCP in Portuguese), name it Autonomous, enter the MCP endpoint above, choose OAuth when requested, scan/verify the tools, authorize the same Autonomous account, and finish creating/enabling the app.

Custom MCP availability and write capabilities depend on the ChatGPT plan and workspace permissions.

5. Or connect Claude manually

In Claude web or Claude Desktop, open Customize → Connectors, click +, choose Add custom connector, name it Autonomous, and paste the same MCP endpoint.

https://mcp.octamundi.com/v2

Click Add, complete the Autonomous sign-in/authorization flow, then make the connector available in your chat from + → Connectors. Automatic tool access is fine for normal use.

6. Connect Gemini

Gemini custom apps can connect to an MCP server URL where that feature is available. In Gemini web, open Settings & help → Connected Apps → Add a custom app, then use the Autonomous MCP endpoint and complete authorization.

https://mcp.octamundi.com/v2

Gemini CLI also supports Streamable HTTP MCP servers. Add Autonomous as a remote HTTP MCP server and complete OAuth when prompted.

Gemini custom-app availability depends on account eligibility, region, language and product surface. Gemini CLI has separate availability and enterprise controls.

7. Connect Grok

Grok supports custom MCP connectors. Open grok.com/connectors, choose New Connector → Custom, enter the Autonomous MCP URL and complete authentication.

https://mcp.octamundi.com/v2

8. Google Antigravity IDE and CLI

Open the Agent panel → ... → MCP Servers → Manage MCP Servers → View raw config. In Antigravity 2.0, open Settings → Customizations → Installed MCP Servers → Open MCP Config.

Merge the following into ~/.gemini/config/mcp_config.json, retaining your other entries under mcpServers. Workspace-specific projects can use .agents/mcp_config.json instead.

{
  "mcpServers": {
    "autonomous": {
      "serverUrl": "https://mcp.octamundi.com/v2"
    }
  }
}

Refresh the MCP servers and select Authenticate. Complete your Autonomous sign-in, then paste the authorization code into Antigravity. The registered OAuth callback is https://antigravity.google/oauth-callback.

Use serverUrl for remote HTTP MCP, not url or httpUrl. Do not paste access tokens into your project, disable authentication, overwrite other MCP entries, or confuse the remote MCP endpoint with the local autonomous command.

For client-specific updates, see the official Google Antigravity MCP documentation.

9. Other supported MCP clients

Autonomous also allows Cursor, GitHub Copilot / VS Code MCP clients, and Sensei. Use the same remote MCP endpoint and the authentication method supported by that client.

Client support can vary by plan, organization policy, product surface and transport/authentication support.

10. Test the complete path

Once the computer is online and the AI client is connected, ask it to list your connected computers. Then give it a real task that needs the computer.

@Autonomous list my connected computers and tell me which one is online.

For security details, see Security. The MCP endpoint always stays https://mcp.octamundi.com/v2.

11. What you can ask Autonomous to do

Give your connected AI a goal in plain language. Autonomous provides authorized tools to work with your computer, files, search results, applications and connected services. Availability depends on the connected device and client.

• Find, read, organize, create and edit files, including supported PDFs, documents and spreadsheets.
• Run development commands, inspect progress, search project files and verify changes.
• Discover devices, choose the computer you want to use and inspect supported operations.
• Use Workloops to organize a multi-step goal and resume unfinished work in a conversation.
• Discover relevant Skills and integrations, including Google Workspace and Apify.

Autonomous, inspect my project, fix the failing tests, and show me the verified result.

12. Workloops: from a goal to verified results

Workloops help your AI organize progress across multiple steps. Ask the AI to create and START a Workloop, carry out each authorized action, check results, and finish only when the outcome has been verified.

Autonomous, create and start a Workloop to audit my website, fix the problems you can safely fix, and verify each step.

Saving a Workloop is not the same as executing it. The calling AI performs the work. Workloops do not independently run an AI after a chat closes; capabilities and continuation depend on your active AI client and permissions.

13. Google Workspace

With Google Workspace connected on your chosen computer, your AI can perform authorized tasks in Gmail, Drive, Calendar, Docs and Sheets.

Autonomous, summarize my latest important Gmail messages and prepare a Google Sheets task list.

Google authorization is required before using private account data. Complete the sign-in and consent privately on your own device. Do not send passwords or authorization codes in chat.

14. Apify: research and public data

Autonomous can consult the public Apify Actor directory without a token when researching topics. If a suitable Actor is available, your AI can use it to collect structured public data with your authorization.

Autonomous, research 100 public technology companies and deliver a CSV with verified sources.

Searching the public Actor directory requires no token. Some Actor runs, private data or paid features require authentication or payment. An Actor listing is not a completed research result; your AI should use other sources when necessary and never silently trigger costs.

15. Agent Skills and guidance

Agent Skills provide reusable instructions for tasks such as research, automation, coding and file handling. The AI can discover relevant installed Skills and load the guidance needed for your goal.

Autonomous, find the relevant Skills for my React project, apply only what is appropriate, and test the outcome.

Skills do not grant extra permissions. Review third-party packages before installation or execution; always verify the actual work produced.

16. Safety, access and troubleshooting

• Your AI may act only with the access granted to the connected account and device. Sensitive or paid actions may require your approval.
• Keep your own computer connected for tasks that require it. If no device appears, open Autonomous on that computer and check its connection.
• When a task fails, ask your AI to inspect the error, try permitted alternatives and report what it could verify.
• If a newly introduced tool is missing, refresh the Autonomous connection in your AI app or start a new conversation.
• Use the Security page for product-level privacy information; never paste private credentials into a prompt.