Everclaw is an agent-focused project; David’s “Morpheus skill” for OpenClaw lets agents call Morpheus inference via a local HTTP endpoint. You provide the endpoint by running a C-Node on the same machine.
This page is a mirror summary, last verified for v7.0.0. For canonical Everclaw guidance, see everclaw.xyz.

​
Pattern

The skill thinks it’s calling a local OpenAI-compatible API. The C-Node handles wallets, sessions, and routing on the back end.

​
Setup steps

1

Run a C-Node

Follow C-Node setup. Bind :8082 to loopback only.
2

Approve some MOR allowance once

See “Approve once” in C-Node setup.
3

Pre-open a session (optional)

Some agent skills want a stable session for the run. Open one and pass the sessionId to the skill via env var.
SESSION=$(curl -s -X POST \
  'http://127.0.0.1:8082/blockchain/models/<modelId>/session' \
  -H 'Authorization: Basic <base64(user:pass)>' \
  -H 'Content-Type: application/json' \
  -d '{"sessionDuration": 3600}' | jq -r .sessionId)
echo $SESSION
4

Configure the skill

Set the skill’s OpenAI-base-URL to http://127.0.0.1:8082 and inject Authorization: Basic <base64(user:pass)> and session_id: <sessionId> headers per request. See your skill’s docs for the exact env names.
5

Validate

Run a sample task in the skill and watch the C-Node logs. You should see prompts being forwarded and the session metering correctly.

​
Permissions

Run agents under a separate proxy.conf user with a method whitelist. For an Everclaw skill that only needs chat completions plus session lifecycle: 
rpcauth=agent:<salt>$<hash>
rpcwhitelist=agent:chat,open_session,close_session,get_balance
rpcwhitelistdefault=0
See API auth for the full method list.

​
Failure modes to plan for

  • Provider unreachable — agent should retry, then the C-Node closes the stuck session and returns the unused MOR.
  • Session expired mid-task — agent should detect 401/403/429 from the C-Node and re-open a session.
  • Allowance exhausted — approve more MOR (see C-Node setup) or set up a top-up cron.