Claude Continuity

You can use the Claude Continuity MCP Server to protect your project context across threads and sessions. It saves your project state locally and restores it automatically, so you don’t lose track when conversations hit token limits. This makes it easy to pick up right where you left off across Claude Desktop and Claude Code.

How to use

Connect your MCP client to the Claude Continuity server and leverage its state persistence from now on. Start the local MCP server and configure your Claude applications to point to it. When you save a project state, the server stores the current context, notes, files touched, and next steps. When you start a new thread or session for the same project, you can load the saved state to continue seamlessly. You can manage multiple projects at once and access quick summaries without loading full details.

How to install

Prerequisites you need to have before installing: Python 3.8 or higher and the MCP SDK. You will also need a client that can connect to MCP servers (such as Claude Desktop or Claude Code) to use the server.

# 1. Clone the project repository to your home directory
cd ~
git clone https://github.com/yourusername/claude-continuity.git

# 2. Install dependencies
cd ~/claude-continuity
pip3 install mcp

# 3. Prepare the Claude Desktop client with MCP server configuration (example shown below)

To configure Claude Desktop, add the following MCP server entry to Claude’s configuration for your user. This example uses a Python-based server path you would customize to your setup.

{
  "mcpServers": {
    "claude-continuity": {
      "command": "/opt/homebrew/bin/python3.11",
      "args": ["/Users/yourusername/claude-continuity/server.py"],
      "env": {}
    }
  }
}

Configure Claude Code to use the server globally:

claude mcp add --scope user claude-continuity python3 ~/claude-continuity/server.py

Additional configuration and startup notes

When you start the server, ensure Python can execute the server script and that the path to server.py is correct for your environment. The server stores data under your home directory in a hidden folder and manages backups automatically.

Data storage and structure

All project states are stored locally in your home directory under a hidden folder. Example layout:

~/.claude_states/
├── project-name-1/
│   ├── state.json
│   ├── summary.txt
│   └── state.backup.*.json
└── project-name-2/
    ├── state.json
    ├── summary.txt
    └── state.backup.*.json

Each saved state includes the project name, the saved state data, a human-readable summary, contextual metadata (files touched, issues, next steps, dependencies), and a timestamp with a version number.

Troubleshooting and tips

If the server isn’t connecting, verify your Python installation is at least 3.8, start the server manually to check logs, and review client logs for errors.

If you encounter permission issues, grant execute permissions to the server script and ensure the path you configured is correct.

If the server isn’t found by the client, double-check the command and arguments you configured in Claude Desktop and Claude Code. Use absolute paths where possible.

Available tools

save_project_state

Save the current project context and state for later restoration.

load_project_state

Restore a previously saved project context and state for seamless continuation.

list_active_projects

Show all projects currently tracked by the MCP server.

get_project_summary

Retrieve a quick overview of saved states without loading full details.

delete_project_state

Remove a saved project state from the server storage with a safety confirmation.

update_project_context

Update contextual metadata for a saved project without altering the full state.