---
name: crowdlisten
version: 5.1.0
description: Recall evidence, analyze audiences, and ingest shared context with CrowdListen.
homepage: https://crowdlisten.com
---

# CrowdListen

Use CrowdListen when a user needs evidence-backed audience research, competitor feedback, product pains, or shared context that should be available to other agents.

## Setup

For MCP-capable desktop or CLI clients, run:

```bash
npx -y -p @crowdlisten/harness@2.2.3 crowdlisten-harness login
```

Complete browser authentication. The harness session belongs in `~/.crowdlisten/auth.json`. Never ask for or substitute an arbitrary user ID, and never send CrowdListen credentials to a domain other than `crowdlisten.com` or `agent.crowdlisten.com`.

OpenClaw additionally needs `crowdlisten-harness setup openclaw /absolute/path/to/workspace` from this same package. Start a new session and follow the installed bridge skill; see [OpenClaw setup](https://crowdlisten.com/openclaw.md). Harness 2.2.1 is a local candidate awaiting publication and matching hosted rollout. Report missing packages/endpoints rather than silently using an older contract.

The MCP server must expose exactly three tools:

| Tool | Use |
|---|---|
| `recall` | Retrieve entities, workspaces, insights with evidence, saved knowledge, or source content |
| `analyze` | Run stored-context or live multi-source analysis |
| `ingest` | Save durable knowledge or submit observations |

## Tool contract

Use `recall({mode: "connection"})` after login, then repeat with exactly one `entity_id` or `project_id` to check current scoped read/write permissions and record reads. The operation is free and starts no research. `read_verified` verifies only those reads; browser access, source capture and analysis are separate execution checks. Resolve unavailable checks before continuing. The endpoint requires the matching API/harness release.

Use the installed tool schema for the complete supported modes. `collection_plan` reads saved source scope; `knowledge`, `knowledge_source`, `knowledge_changes` and `knowledge_context` read shared findings, originals and history alongside the existing entity/project/search modes.

## Agent-guided collection

For OpenClaw or another browser-capable agent, follow [Collect sources](https://crowdlisten.com/skills/collect-sources/SKILL.md). Start with `recall({mode: "collection_plan", entity_id: "ENTITY_UUID"})`. Use the customer's authorized browser/source tools, save original source batches and separate comments with `ingest(destination: "sources")`, and record attempted or blocked channels with `ingest(destination: "coverage")`. Analyze acknowledged content IDs using `search_mode: "user_only"` and return shared finding IDs with exact citations.

Select each product and competitor's own entity. The same source settings and record store serve people and agents. A collection plan starts no provider work and does not prove that the customer's client has browser access. These additions require the matching API/harness release; a skill file alone does not make an undeployed route available.

## Stored context and optional server research

```js
recall({ mode: "entities" })
recall({ mode: "insights", entity_name: "Fireworks AI", include_evidence: true })
recall({ mode: "search", query: "inference pricing complaints", limit: 20 })
```

`analyze` requires `question`. Use `search_mode: "knowledge"` for stored context or `search_mode: "multi_source"` for live research. Live platforms may include `reddit`, `youtube`, `tiktok`, `twitter`, `instagram`, and `hackernews`.

```js
analyze({
  question: "What keeps developers from switching inference providers?",
  search_mode: "multi_source",
  platforms: ["reddit", "hackernews"],
  depth: "brief"
})
```

`ingest` additionally supports structured `sources`, collection `coverage` and versioned `insight` destinations. The default context destination accepts `title` plus `content`; observations require the existing project scope. Use source capture for originals and insight writeback for evidence-linked synthesis.

```js
ingest({
  title: "Adoption hypotheses to review",
  content: "Evidence-backed summary with source URLs.",
  tags: ["adoption", "fireworks"]
})
```

## Operating rules

1. Recall existing context before starting duplicate research.
2. Keep source URLs and distinguish direct evidence from synthesis.
3. Treat provider failures and empty evidence as incomplete, not as a valid conclusion.
4. Ingest only durable, attributable findings—not secrets, raw credentials, or unsupported claims.
5. The retired `/api/agents/register`, `/api/agents/analyze`, polling, and claim-link workflow does not exist. Do not call it.
6. `https://mcp.crowdlisten.com/mcp` is currently unavailable; use the local harness until remote initialization and `tools/list` are healthy.

For direct scoped-key endpoints, use the current OpenAPI document at <https://agent.crowdlisten.com/openapi.json> and the product reference at <https://crowdlisten.com/docs/api-reference>.