# Connect OpenClaw to Crowdlisten

Use Crowdlisten as the shared record store and OpenClaw's authorized browser/source tools to collect evidence. Select your product or competitor in Crowdlisten, configure channels and tracked accounts in Sources settings, then use **Save and collect with my agent**.

This setup targets harness **2.2.1**, its matching backend and source/knowledge migrations. It is a local release candidate; 2.2.1 is not yet published and the hosted connection endpoint has not yet been updated. Do not distribute these commands as working production onboarding until those releases are verified.

## Install on the agent's host

Use Node.js 22 or newer with npm/npx and an existing OpenClaw workspace. The bridge is pinned to mcporter 0.7.3, compatible with Node 22. A newer mcporter release may require a newer Node runtime. No global bridge installation is needed.

```sh
npx -y -p @crowdlisten/harness@2.2.1 crowdlisten-harness login
npx -y -p @crowdlisten/harness@2.2.1 crowdlisten-harness setup openclaw /absolute/path/to/workspace
```

Use the current agent's workspace, especially for multiple agents or profiles. `openclaw config get agents.defaults.workspace` identifies a configured default; check the actual agent's configuration before applying it to another agent. Pass an existing absolute path, quoted when it contains spaces. Setup never guesses the workspace.

The command writes `skills/crowdlisten/{SKILL.md,COLLECTION.md,mcp.json}` plus an ownership manifest. It preserves OpenClaw account/provider settings and other MCP servers. Repeating setup updates unchanged managed files; customized, unowned, linked or interrupted installations require review and are left intact. The response says `configured_unverified` until you perform the checks below.

No access token or provider credential goes in the installed skill/configuration. The stdio server uses the same host user's browser-login session. A sandbox, remote gateway or another OS account needs its own authorized connection and executable dependencies. Installing a local skill does not connect every sandbox.

## Verify before collecting

Start a new OpenClaw session and ask it to read the `crowdlisten` skill. `openclaw skills info crowdlisten` confirms discovery only. Run its schema and free connection checks:

```sh
npx -y mcporter@0.7.3 --config "/absolute/path/to/workspace/skills/crowdlisten/mcp.json" list crowdlisten --schema --output json
npx -y mcporter@0.7.3 --config "/absolute/path/to/workspace/skills/crowdlisten/mcp.json" call crowdlisten.recall --args '{"mode":"connection"}' --output json
```

Exactly three tools should be listed: `recall`, `analyze`, `ingest`. Select an authorized entity using recall mode `entities`, then repeat the connection check with that ID. Read permission/read-check failures. A 404 means the backend needs its matching release; package-not-found means the candidate has not shipped. Do not substitute an older tool contract silently.

## Collect and retain the result

Paste the existing Sources handoff into OpenClaw. The installed COLLECTION.md describes the same workflow as other clients:

1. Read `collection_plan` for the chosen product or competitor.
2. Use authorized browser/source tools for the saved channels and bounds.
3. Capture originals and separate comments through ingest destination `sources`; retain returned IDs. Record sampled/empty/blocked coverage separately.
4. Analyze acknowledged IDs with `search_mode: "user_only"`. Preserve durable job/request IDs across disconnects.
5. Read canonical findings and exact originals; another authorized client must retrieve the same record/revision.

For comparison, repeat within each product and competitor entity rather than moving evidence between them. Keep customer experience distinct from marketing, unverified commentary and generated interpretation. Browser blocks are coverage gaps, not permission to bypass access controls.

Bridge checks use controlled local auth/API responses. Earlier API/PostgreSQL checks exercise the record pipeline. Neither proves live OpenClaw browser collection or a production customer journey; those remain release gates.