> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.beatapi.io/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.beatapi.io/_mcp/server.

# How can an assistant or workflow retrieve actual social data?

> Connect AI clients to BeatAPI via remote MCP at https://beatapi.io/mcp or send fixed HTTP workflows to POST https://api.beatapi.io/v1/social-data/call for verifiable social data.

An assistant retrieves actual social data when the assistant's client can call BeatAPI tools: connect the remote MCP server `https://beatapi.io/mcp` and check that `capabilities_search`, `capabilities_inspect` and `capabilities_run` appear. A workflow retrieves actual data when an HTTP step POSTs to `https://api.beatapi.io/v1/social-data/call`. Use a fixed HTTP workflow when the steps never change, and let the assistant choose tools when the questions vary.

For example, you want to select installation tutorials from a WeChat Official Account, then see what questions users ask on Xiaohongshu and Douyin. Check what your client can execute first, then choose the HTTP workflow or the assistant-selected tools.

## Verify the client’s tools

1. **Add the server.** If your client supports remote HTTP MCP and custom headers, set the server URL to `https://beatapi.io/mcp`, and store `Authorization: Bearer <BEATAPI_API_KEY>` as a client credential.
2. **Check the tool list.** Confirm that the client lists `capabilities_search`, `capabilities_inspect` and `capabilities_run`.
3. **Run a first check.** Ask the assistant to find one action and inspect the action's inputs, then request the agreed first batch.
4. **Keep the record.** Save the action, status, request ID and the material actually read.

How CLI assistants, desktop extensions and hosted workflows connect depends on the client version, header support and network settings. If the connection fails, check tool discovery first, then check authentication, transport and execution separately. BeatAPI's public reading actions use a BeatAPI key. An existing platform plugin manages its own account session and data source.

## Use HTTP for a fixed sequence

In an environment with an HTTP step, POST to `https://api.beatapi.io/v1/social-data/call` with `application/json` and the BeatAPI key from the credential store. Send one target first and verify the response before adding selection and export steps.

```json
{"action":"xiaohongshu.app_v2.search_notes","params":{"keyword":"咖啡机 清洁","page":1}}
```

To list a verified Official Account's articles, use a separate step:

```json
{"action":"wechat_mp.v2.fetch_account_articles","params":{"username":"<verified_username>","offset":"","item_show_type":0,"raw":false}}
```

Pass only the selected article URLs to the article body step, and keep the original links and the selection reasons. If you already have saved Xiaohongshu or WeChat responses, go directly to selection and export. See [Xiaohongshu table handoff](/guides/xiaohongshu-note-workflow-handoff) and [WeChat article inventory and statistics](/guides/wechat-python-article-review).

## Choose an entry point for your environment

| Environment                                            | Check first                                                                        | First delivery                                                        |
| ------------------------------------------------------ | ---------------------------------------------------------------------------------- | --------------------------------------------------------------------- |
| Tool clients such as Claude Code, Cursor or Codex      | Remote HTTP MCP, auth headers and tool discovery                                   | A sourced first batch of works or articles                            |
| Workflows such as Dify or Coze                         | Configurable HTTP for fixed steps, or verified MCP support for dynamic tool choice | Input, original response and selected links before requesting details |
| Desktop clients such as WorkBuddy, Doubao or Qwen Work | Whether the installed connector accepts the endpoint and headers                   | Confirmed tool execution and response before expanding                |
| A client using DeepSeek or another model               | Whether the host can make tool calls or provide saved material                     | Original evidence distinguished from model summaries                  |
| Chat-only client                                       | Provide saved content and its scope                                                | Sourced summaries, categories and gaps                                |

BeatAPI defines the MCP endpoint, HTTP endpoint, and tool schemas. Each client's connection, templates and publishing still need checking in your own environment. An existing skill can store the task steps and reporting requirements; check the execution record, the response and the sources anyway.

Prompt for a connected assistant:

> Use this verified Official Account identifier to fetch a batch of articles and read two about coffee machine cleaning. Find a batch of related Xiaohongshu and Douyin content. Deliver original evidence, material types, request count and what still needs reading.

To read more of the same saved result, see [result reading](/faq/agent-result-reading). New pages and details are additional queries; see [pricing](https://beatapi.io/pricing).

For example, if your weekly brief reads the same accounts every week, use a fixed HTTP workflow that saves the accounts and their list pages. When an editor asks about an unexpected note, let a tool-enabled assistant choose the action. If the client cannot connect, give the model last week's saved lists and notes, and label the collection date in the brief.

## Check the input contract when handing off to an analysis tool

If an analysis tool such as BettaFish shows no data, check whether the tool reads files, database tables or tool calls.

1. Save a small public Xiaohongshu and Douyin sample, keeping the original responses.
2. Map the work IDs, text scope, date units and comment links that the installed tool version requires.
3. Verify one saved record end to end, from import to query.

Check BettaFish's actual schema and connection in your own installation. The BeatAPI workflow supplies traceable input files and does not write into BettaFish's database.

[Read social sources in Dify](/guides/dify-social-research) · [Social lookups in Coze](/guides/coze-social-research)

[Check BettaFish data and verify sample read-back](/guides/bettafish-social-data)

[Configure each of the three desktop clients](/guides/desktop-social-research-mcp)

* [Social data MCP in office clients](/guides/office-clients-social-mcp)
* [DeepSeek tools and social evidence](/guides/deepseek-social-research)