Skip to content

On this page you will connect MMW to Claude Code, verify the connection and, if you want, turn on session history upload.

  • The remember, search, forget, validate_memory and gateway_call tools in every Claude Code session.
  • The read-only mmw_sync_status tool, when you connect through the local agent.
  • Claude Code session history upload, only after you explicitly consent in a terminal.
  • The same memory that Cursor, Codex and claude.ai see with the same key.
  • An MMW API key (starts with mmw_); see the Quickstart.
  • Claude Code with the claude command in your terminal.
  • For the local agent: Python 3.11+ and uv (the uvx command).
  1. Run in a terminal:

    Terminal window
    claude mcp add mmw -s user \
    -e MMW_API_KEY=mmw_your_key \
    -e MMW_ENDPOINT=https://mcp.mmwhub.tech \
    -- uvx --from https://app.mmwhub.tech/downloads/mmw-agent/mmw_agent-0.1.4-py3-none-any.whl mmw-agent

    The -s user flag makes MMW available in every project. Without it the server is added for the current directory only.

  2. Restart Claude Code if it was open.

  3. Verify the connection (see below).

The command writes an mcpServers.mmw block to ~/.claude.json. You can also add it by hand:

~/.claude.json
{
"mcpServers": {
"mmw": {
"command": "uvx",
"args": ["--from", "https://app.mmwhub.tech/downloads/mmw-agent/mmw_agent-0.1.4-py3-none-any.whl", "mmw-agent"],
"env": {
"MMW_API_KEY": "mmw_your_key",
"MMW_ENDPOINT": "https://mcp.mmwhub.tech"
}
}
}
}
  1. In a terminal run claude mcp list: the mmw server should be listed as connected.

  2. Inside a Claude Code session type /mcp: the mmw server shows remember, search, forget, validate_memory and gateway_call, plus mmw_sync_status when connected through the agent.

  3. Ask:

    Remember in MMW: staging is deployed with the systemd unit app-staging. Fact key: deploy-staging.

    Claude Code calls remember and reports the record ID with the unverified status.

  4. Open a new session and ask:

    How do we deploy staging? Check MMW.

    Expected answer: the assistant calls search and replies that staging is deployed with app-staging, citing the MMW record.

You can check the agent on its own, without Claude Code:

Terminal window
MMW_API_KEY=mmw_your_key MMW_ENDPOINT=https://mcp.mmwhub.tech \
uvx --from https://app.mmwhub.tech/downloads/mmw-agent/mmw_agent-0.1.4-py3-none-any.whl mmw-agent status

The agent can upload your Claude Code history to MMW, but only after a person confirms consent in an interactive terminal. A model or a script cannot grant consent.

Terminal window
mmw-agent consent # shows how many sessions it found and asks for permission
mmw-agent revoke # withdraw consent: uploads stop immediately

If mmw-agent is not installed globally, run it as uvx --from <wheel URL> mmw-agent consent with the same MMW_API_KEY and MMW_ENDPOINT.

User and assistant messages, tool calls and tool results are uploaded; model reasoning, system prompts, file snapshots and paths are not. Secrets are removed on your computer before sending and once more on the server. See Session history.

For other errors see Errors and Agent troubleshooting.