Innernet

What innernet does

The innernet MCP server gives connected AI tools access to one shared, versioned memory for your projects and selected personal information. It is designed to preserve what you are building, decisions and their rationale, active work, and task status across different AI clients.

The repository packages a remote connector and two instruction skills. The main skill helps an assistant identify the relevant project, retrieve useful context, answer from that context, save decisions, maintain the project task list, and retain durable personal details that you explicitly choose to share. The handoff skill supports session continuity by retrieving where work stopped at the beginning and recording a concise handoff at the end.

This places innernet in the Knowledge & Memory category. It is a fit for projects where conversations happen across multiple hosts or AI applications and the next session needs more than the current chat history.

How it works

The connector points clients to https://innernet.live/api/mcp, which is also listed in the official MCP registry as live.innernet/innernet. Authentication happens once through innernet's OAuth page in a browser. A free innernet.live account is required.

The plugin does not run scripts, hooks, or a local server. When an assistant makes a connector call, the remote service receives the information carried by that call, such as a project name, a question, or text to save. The service returns the matching memory. It does not read the host assistant's memory, chat history, or files.

Changes are stored in the innernet account with version history and can be edited or deleted from the dashboard. Data-changing operations require confirmation, and the supplied skills instruct the assistant to report what it saved. Personal memory uses disclosure levels; private facts are not returned to an AI.

Setup and configuration

Install the package through the relevant client mechanism or configure any MCP client with the remote endpoint. Claude.ai and Cowork use the plugin's connector settings. Claude Code and Codex provide marketplace commands, Cursor supports its marketplace or repository-based plugin installation, and Gemini CLI installs the repository as an extension. Other MCP clients can add the remote server directly.

After installation, connect the innernet MCP server and complete the browser sign-in. Do not paste an API key into a chat. The repository contains client-specific manifests for Claude, Codex, Cursor, and Gemini CLI, plus shared MCP and skill configuration.

Tools and capabilities

The available workflow supports:

• Loading context for the project under discussion.
• Answering questions from the relevant stored memory.
• Saving decisions and other requested project information.
• Keeping project tasks current.
• Storing durable personal information subject to disclosure settings.
• Retrieving a previous stopping point and saving a session handoff.

The README describes these capabilities through skills rather than listing individual MCP tool names. Use natural requests such as asking to load project context, check a past decision, find the last stopping point, or save progress.

Limitations and notes

The innernet MCP server depends on the hosted innernet.live service and an account; the repository does not provide a local fallback. Only information included in connector calls is sent to the service, and the plugin does not inspect local files or conversation history automatically. You remain responsible for deciding what personal information to disclose.

The repository is MIT licensed. Its privacy policy and terms are published at innernet.live, and support documentation is available at the service's connect documentation page.