增强 · 实时数据LinkedIn Agent

LinkedIn 数据 Skill

通过 UnifAPI MCP 读取 LinkedIn 公开数据:用于 B2B 研究的 linkedin/... 操作与响应结构。绝不加好友或发消息。

Agent 原生

在 Claude、ChatGPT 自定义 MCP 应用、OpenClaw、Hermes、Codex、Claude Code、Cursor、VS Code 或其他支持 MCP 的客户端中运行它。无需专门的 GUI 流程,也无需单独的 LLM API 密钥。

由实时公开数据支撑

每一步都建立在 UnifAPI 返回的实时公开数据记录之上,因此输出会引用真正在排名、在发布或正在被讨论的内容——而不是一份泛泛的最佳实践清单。

可组合且开源

Skills 相互交叉引用,并存放在一个公开的、MIT 风格的仓库中。在 GitHub 上阅读完整的 SKILL.md、fork 它,或原样在你的 Agent 中运行它。

运行提示词

把它粘贴进 Codex 或 Claude Code

这段提示词有意可自由编辑。替换其中的用户名、市场、预算与活动目标,然后让 Agent 在需要实时公开数据时调用 UnifAPI MCP。

Using the LinkedIn Data Skill, research stripe on public LinkedIn: pull the company profile and recent posts, run a company-filtered job search for one relevant keyword, and return a sourced brief with unsupported capabilities called out. Read-only.
这个 skill 如何运作

由 SKILL.md 渲染出的完整 skill

Use this skill to read the eight LinkedIn operations currently published by UnifAPI. Treat this list as the capability boundary: do not invent legacy sub-resources or promise people search, keyword post search, company employee listings, reactions, follower counts, or job counts.

Use UnifAPI for live evidence

Connect with the unifapi skill, inspect the chosen operation with get_operation, then call it with call_api. Successful responses use UnifAPI's standard envelope:

{
  "data": {},
  "pagination": { "has_more": false, "next_cursor": null },
  "billing": { "records_charged": 1 }
}

Keep billing in cost reporting. For list operations, pass pagination.next_cursor as the next request's cursor only when pagination.has_more is true.

Core operations

NeedOperation
Company profilelinkedin/companies/{slug}
Company postslinkedin/companies/{slug}/posts
Member profilelinkedin/users/{username}
Member postslinkedin/users/{username}/posts
Search jobslinkedin/search/jobs
Job detaillinkedin/jobs/{id}
Post detaillinkedin/posts/{id}
Post commentslinkedin/posts/{id}/comments

Workflow

  1. Resolve the public key. Read a company {slug} from /company/<slug> or a member {username} from /in/<username>.
  2. Inspect the contract. Call get_operation before the first live call so query fields, pagination, response shape, and billing are current.
  3. Read the narrowest record. Start with the company or member profile; add its posts only when the task needs current themes or authored evidence.
  4. Research hiring carefully. Read company_id from the company profile, then call linkedin/search/jobs with a meaningful keyword and the numeric company filter. This is a filtered search, not an exhaustive company job list or a historical job-count series.
  5. Open evidence. Use a returned job id with linkedin/jobs/{id} or a post id with linkedin/posts/{id}; fetch comments only when reception matters.
  6. Report limits. Separate fields read directly from hypotheses inferred from posts, jobs, or company size. State when a requested capability is outside the current eight-operation surface.

Shape notes

  • Company profiles expose the URL slug plus company_id, description, industry, size/headcount fields, headquarters, specialties, and public counts when present.
  • Member profiles expose the URL username, headline, location, website, and public status flags. Follower and connection counts are not part of the current profile contract.
  • Job search is cursor-paginated and requires a free-text keyword. The company filter accepts the numeric company_id from a company profile.
  • Post lists and comments are cursor-paginated. Preserve returned ids and source URLs so every downstream claim is traceable.

Gotchas

  • Companies are keyed by URL slug; members are keyed by URL username. Do not substitute display names.
  • A company profile's headcount is a current public field, not proof of growth over time.
  • A filtered job result is evidence that a matching role was returned, not proof of the company's complete hiring plan.
  • Capacity-limited collection and job operations may retry inside the gateway. Do not add a second retry loop unless the returned error contract says it is safe.
  • If an operation is absent from list_operations, treat it as unavailable rather than guessing an older path.

Output

Return the records the caller needs with source URLs, dates where available, a clear distinction between observed facts and inference, and one cost line using billing.records_charged.

  • linkedin-account-research — turns company profiles, posts, filtered job results, named member profiles, and news into an account brief.
  • account-news-signals — corroborates dated company news against official company posts.
  • buying-signal-monitor — combines X intent search with named-account LinkedIn evidence.
  • unifapi — connects MCP and exposes the current operation catalog and schemas.

源码:GitHub 上的 linkedin/SKILL.md——可在那里提交 PR 来改进它。

公开数据工具

这个 skill 调用的实时 API

这个 skill 提到的每个操作都属于这些 UnifAPI 平台之一——它们对于产品代码、调试与自定义 Agent 流程仍然可见、可调用。

  • 针对每个常见 B2B 读取任务的具体 `linkedin/...` 操作
  • 面向公司、用户、职位与帖子的响应契约与结构说明
  • 确定性配方:客户调研、定向招聘证据、会员、帖子
  • 平台注意事项(slug 与 username 键、游标分页、计费)