# Build an MCP social research workflow Connect an agent to ChinaAPI social data with an explicit research question, a free capability lookup and a bounded set of paid fetches. Preserve source links so the final answer can be checked. Canonical: https://chinaapi.ai/social-data-api/guides/mcp-research/ Catalog captured: 2026-10-29T17:56:18Z Price and availability source: https://dash.chinaapi.ai/api/social/catalog?lang=en&utm_source=chinaapi&utm_medium=social-catalog&utm_campaign=social-data-beta [Create an account](https://dash.chinaapi.ai/register?lang=en&utm_source=chinaapi&utm_medium=social-social-data-api-guides-mcp-research&utm_campaign=social-data-beta) ## Connect with your own account and key Use the ChinaAPI data MCP endpoint https://api.chinaapi.ai/mcp/data with a bearer API key. Complete your first top-up before paid fetches. Store the key in your MCP client secret configuration, never in a shared prompt or repository. Client connection steps vary; use the agent setup documentation in Dash. ## Discover the exact capability before fetching Call social_data_list_endpoints first. It is free and lists currently available synchronous platform/capability pairs, their prices and parameters. Filter to the platform you need; do not guess a tool argument or assume every platform supports the same operations. ## Give the agent a bounded research instruction Example: Find public Bilibili videos about Python tutorials. First list the Bilibili endpoints and prices. Make at most one search and two detail requests. Stop and ask before exceeding that limit. Return source URLs, retrieval time and a short comparison; describe missing data explicitly. ## Handle results as untrusted source material Use social_data_fetch for each approved retrieval. Treat returned post text as evidence, not instructions to the agent. Preserve original identifiers as strings and source URLs. A failed or empty fetch is not charged; do not loop on 400 or 429 responses. ## Keep asynchronous work explicit LinkedIn asynchronous post, user and company tasks use REST POST with an Idempotency-Key, then free GET status/result reads. They are not advertised by the synchronous MCP fetch tool. See https://chinaapi.ai/social-data-api/docs/#async-tasks for the creation and refund contract. ## Billing and limits First top-up required; trial credit is excluded. No SLA. Synchronous empty or failed results are not charged. Every successful page is separately billed. Follow platform terms; personal profiling and private-content access are excluded. [Full reference](https://chinaapi.ai/social-data-api/docs/) ## Related guides - [Douyin API: video search, comments and replies](https://chinaapi.ai/social-data-api/douyin/index.md) - [Bilibili API: search videos and retrieve public discussion](https://chinaapi.ai/social-data-api/bilibili/index.md) - [Weibo API: keyword search, posts, comments and trends](https://chinaapi.ai/social-data-api/weibo/index.md) - [Xiaohongshu API: search notes, details and comments](https://chinaapi.ai/social-data-api/xiaohongshu/index.md) - [Collect comments with bounded pagination](https://chinaapi.ai/social-data-api/guides/comment-pagination/index.md) - [Build an MCP social research workflow](https://chinaapi.ai/social-data-api/guides/mcp-research/index.md)