Integrates AgentKit into an always-on host so OpenClaw or Hermes can act as a named user. Use when the user wants OpenClaw, Hermes, clawhub install, or an always-on host skill. It does not write AgentKit in app code (that's `integrate-agentkit`) or expose tools over MCP (that's `expose-agentkit-mcp`).
Installs into .claude/skills of the current project.
Are you the author of Integrate Agentkit Host?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/scalekit-inc-integrate-agentkit-host)
---
name: integrate-agentkit-host
description: >
Integrates AgentKit into an always-on host
so OpenClaw or Hermes can act as a named user.
Use when the user wants OpenClaw, Hermes,
clawhub install, or an always-on host skill.
It does not write AgentKit in app code
(that's `integrate-agentkit`)
or expose tools over MCP
(that's `expose-agentkit-mcp`).
---
# Integrate AgentKit Host
Wire an always-on **host** so it can act as a named user through AgentKit. Then stop.
A **host** is a long-lived agent process (OpenClaw or Hermes). It is not this git repo.
## Guardrails
- **MUST** wait for dashboard credential values. **MUST NOT** invent them.
- **MUST** pass the exact dashboard Connection Name. Never invent a slug.
- **MUST** put only Scalekit client credentials on the host. Provider tokens stay in Scalekit.
## Gotchas
- Read `SCALEKIT_ENVIRONMENT_URL`, `SCALEKIT_CLIENT_ID`, `SCALEKIT_CLIENT_SECRET`, and `SCALEKIT_IDENTIFIER`. OpenClaw also accepts `TOOL_*` aliases. Never `SCALEKIT_ENV_URL`.
- A **connection** already exists from `setup-agentkit`. This skill starts there.
- The Slack gateway bot is a channel. It is not the named user.
- Do not run `hermes mcp login` against Scalekit. Do not mint a Virtual MCP session token.
- Default host is OpenClaw. Hermes is [references/hermes.md](references/hermes.md).
## Step 1 — Pick the host
- OpenClaw → stay here.
- Hermes → open [references/hermes.md](references/hermes.md). Stop reading this file.
- App code in this repo → name `integrate-agentkit`. Stop.
- Per-user MCP URL → name `expose-agentkit-mcp`. Stop.
If env is missing, collect the three Scalekit client values from [app.scalekit.com](https://app.scalekit.com) → Developers → Settings → API Credentials. Ask for the named-user identifier (email or user id). Do not invent values.
**Done when:** the host is OpenClaw, and the four names exist.
## Step 2 — Confirm the connection
Use the Connection Name already recorded by `setup-agentkit`.
If none is recorded: Gmail is `gmail` when the dashboard has no Gmail row. Any other connector needs a dashboard row. Else name `setup-agentkit` and stop.
**Done when:** a Connection Name is written down.
## Step 3 — Install the OpenClaw skill
```bash
clawhub install scalekit-agent-auth
```
That slug is the shipped ClawHub listing. Do not create a new GitHub repo for it.
**Done when:** `clawhub` lists `scalekit-agent-auth`.
## Step 4 — Write host env
Put Scalekit **client** credentials in the OpenClaw env file (`.env` next to the skill, or the host env). These are not Gmail or Slack tokens.
```bash
SCALEKIT_ENVIRONMENT_URL=https://your-env.scalekit.cloud
SCALEKIT_CLIENT_ID=<from dashboard>
SCALEKIT_CLIENT_SECRET=<from dashboard>
SCALEKIT_IDENTIFIER=<named user>
```
`TOOL_ENV_URL`, `TOOL_CLIENT_ID`, `TOOL_CLIENT_SECRET`, and `TOOL_IDENTIFIER` also work.
From this skill directory:
```bash
cd scripts && uv sync
uv run tool_exec.py --list-connections --provider GMAIL
```
**Done when:** the env file has those names, and `--list-connections` returns a completed connection or a clear empty list.
## Step 5 — Authorize if needed
```bash
uv run tool_exec.py --generate-link --connection-name <CONNECTION_NAME>
```
OAuth: if not `ACTIVE`, show the magic link and wait. API key: if missing or not `ACTIVE`, send the user to the Dashboard and stop.
**Done when:** the connected account is `ACTIVE`, or the next action is on screen.
## Step 6 — Stop
Name one real host prompt for the user, for example “Show my latest unread emails.” Do not write app-code SDK calls.
The host follows [references/runtime-skill.md](references/runtime-skill.md) at chat time.
**Done when:** OpenClaw is installed, env is set, and this skill has stopped.
## Reach for
- `setup-agentkit` if the connection or env is missing
- `integrate-agentkit` for AgentKit in app code
- `expose-agentkit-mcp` for a per-user MCP URL
- [references/hermes.md](references/hermes.md) for Hermes
- [references/runtime-skill.md](references/runtime-skill.md) for the host chat loop
## Live lookups
- Docs index: https://docs.scalekit.com/llms.txt
- Connector catalog: https://docs.scalekit.com/agentkit/connectors.md
- OpenClaw skill: https://github.com/scalekit-inc/openclaw-skill
- ClawHub: https://clawhub.dev/skills/scalekit-agent-auth
- MCP: https://mcp.scalekit.com