Repair

Troubleshooting

Fix the common setup issues: daemon not running, MCP not connected, missing Claude commands, stale context, and support escalation.

Origin teamUpdated May 15, 20265 min read

At a glance

Start with /init in Claude Code, origin doctor in the terminal, or the doctor tool in an MCP client.

Most issues are one of three things: daemon not reachable, MCP server not configured, or the client needs a restart.

Claude Code commands are missing

If /brief, /capture, or /handoff do not appear after installing the plugin, restart Claude Code once.

Then run /init. It should verify the plugin, daemon, MCP route, and memory round trip.

Claude Code

/init

Daemon is not reachable

Origin's daemon listens locally on port 7878. If a client cannot reach it, use /init from Claude Code or the doctor tool from an MCP client to check the daemon state.

For non-Claude clients, rerun npx -y @7xuanlu/origin setup if the local runtime was never installed or status verification failed.

Only one daemon should own the local database. If you have been developing Origin locally, make sure an old daemon from another checkout is not still running.

MCP client is not connected

Check that you ran npx -y @7xuanlu/origin setup first, then configured the client with command npx and args -y plus origin-mcp. Some clients require a full restart after changing MCP settings.

If the client shows the server but tools fail, run doctor. It reports whether the daemon is reachable and how setup is configured.

MCP config

{
  "mcpServers": {
    "origin": {
      "command": "npx",
      "args": ["-y", "origin-mcp"]
    }
  }
}

Recall returns weak context

First make the recall query more specific. Include project names, feature names, decisions, people, or the exact failure mode.

If the memory exists but is stale or duplicated, use /review for pending queues and /distill when a topic deserves a clearer page.

When to open an issue

If /init or doctor reports a daemon, MCP, or install failure you cannot resolve, open a GitHub issue with the client name, command you ran, and the diagnostic output.

Avoid pasting private memory contents into public issues. Describe the setup failure and redact project-specific details.

Open a GitHub issue