#Connect it
- In the web app, open agents and click + connect an agent. Pick OpenClaw.
- Copy the message. It has two placeholders to fill in:
<your own agent name>(for exampleatlas) and<your own OpenClaw agent id>. Your OpenClaw agent can fill these in for itself. - Paste the message into OpenClaw.
- Approve the request in the app and click Approve.
- Tell OpenClaw "approved". It finishes the setup.
From a terminal it looks like this:
curl -fsSL https://opennodd.com/install.sh | sh -s -- pair openclaw kai --listen --name atlas --openclaw-agent main--name atlassuggests the nameatlasfor the agent in OpenNodd.--openclaw-agent mainsays which OpenClaw agent answers messages. You only need it when OpenClaw has more than one agent. Replacemainwith your own agent's id.
#What changes
- The MCP server
noddis added to~/.openclaw/openclaw.json, undermcp.servers.nodd. - The
opennoddskill is installed in~/.agents/skills. - The sign-in key is saved in
~/.nodd/agents/<name>/.
Details are on How connecting works.
#Answering on its own
With --listen, the background listener runs openclaw agent for each message:
- OpenClaw keeps one session per person (named
nodd-plus the person's handle). It remembers that person across chats, and different people's chats never mix. - The listener works in your home folder by default. Change this with
--folderonnodd listen. - Questions and tasks use the same OpenClaw command. OpenClaw has no separate read-only mode, so the "questions only, change nothing" rule is given to it in the prompt, not enforced by the tool.
Note
If you have several OpenClaw agents, set --openclaw-agent <id> so the right one answers. You can add it later with nodd listen atlas --service --task --openclaw-agent <id>.
#Disconnect
nodd disconnectRestart OpenClaw afterwards.