# OpenClaw integration
> Use OpenClaw with Omentir over MCP or REST and a revocable API key for LinkedIn lead discovery, Steal Customers, and outreach operations from a custom agent stack.
- HTML: https://omentir.com/integrations/openclaw
- Markdown: https://omentir.com/integrations/openclaw.md

Point OpenClaw at Omentir's agent APIs with a Bearer token and run sales operator workflows under guardrails.

## Highlights

- Local operator
- Bearer token
- Same REST and MCP contracts

## Setup

1. **Create a token.** Make an Omentir API key and put it in OpenClaw's secret storage, never in a skill file you commit.
2. **Call MCP or REST.** Use the hosted MCP endpoint or the REST routes under /api/agent/v1 with Authorization Bearer.
3. **Paste the operator prompt.** The /for-agents prompt tells OpenClaw to read context first and not to invent billing or signup flows.
4. **Watch the first writes.** List agents after any create or update. Confirm the dashboard matches what OpenClaw claims it did.

## Why OpenClaw users connect Omentir

OpenClaw-style agent stacks want tools that do real work. Omentir supplies hosted MCP and REST tools for product context, lead finders, Steal Customers, leads, schedules, and constrained conversation actions so your agent is not inventing a LinkedIn client from scratch.

## Auth model

Create a workspace API key on the Omentir API keys page. Store it as a secret. Send Authorization: Bearer on MCP or REST calls. Never embed the token in public prompts or git commits. Rotate keys if a runner environment is compromised.

- MCP: https://omentir.com/api/agent/v1/mcp
- REST: https://omentir.com/api/agent/v1
- OpenAPI: https://omentir.com/api/agent/v1/openapi.json
- Guide: https://omentir.com/agents.md

## Recommended runbooks

Fetch agents.md first. Call get_context. Refuse to create agents until the human confirms targeting. On empty lead lists, check list_activity before claiming discovery worked. Reply tools only on existing threads with draft approval.

## Risks unique to autonomous runners

Background agents can loop. Cap retries, require approvals for create and delete, and log every outbound-affecting tool call. Autonomy without an audit trail is how silent targeting drift happens.

## Running OpenClaw against a live workspace

OpenClaw is useful when you want an operator that already lives on your machine. That also means a leaked token is a leaked workspace. Rotate the key if the machine is shared, and start with read-only habits: context, stats, lead lists.

When you let it create a finder, give it a file or message with the ICP. Do not ask it to crawl your whole disk for 'ideas.' The workspace should change because you asked, not because the agent got bored.

## When OpenClaw is the wrong layer

If you just want to talk to Omentir from a hosted chat app, use Claude, ChatGPT, or Grok. OpenClaw adds a local runtime you have to operate. That is worth it for people who already run it. It is extra surface for everyone else.

## Related

- [For AI Agents](https://omentir.com/for-agents.md): Operator prompt and connect overview.
- [OpenClaw LinkedIn leads](https://omentir.com/blogs/openclaw-linkedin-leads.md): Narrative on OpenClaw lead workflows.
- [OpenClaw outreach flows](https://omentir.com/blogs/openclaw-outreach-flows.md): Outreach-oriented OpenClaw reading.
- [REST API integration](https://omentir.com/integrations/rest-api.md): HTTP surface details.

## Frequently asked questions

**Is OpenClaw a first-party Omentir product?**

No. OpenClaw is an external agent stack. Omentir exposes APIs that OpenClaw can call like any other operator client.

**Chat connector or API key?**

Use a revocable API key for custom runners. Chat connector OAuth is for consumer chat apps with that UI.

**Related reading?**

See OpenClaw LinkedIn leads and OpenClaw outreach flows on the blog, plus /for-agents for the operator prompt.