linkedin-post-analytics

Gate (run first)

1. Pattern: CONNECTORS.md three-step for this integration: Discover — no standard LinkedIn MCP here; Configure — user logged into LinkedIn in the host browser (they sign in); Browser access — ask Claude in Chrome and/or computer use (neither / one / both). If neither or no tooling, stop—do not scrape without read-only approval for the chosen mode(s).
2. {user_dir}: Resolve the job-search folder; tracker is {user_dir}/CareerNavigator/tracker.json.
3. Scheduled unattended runs: If this invocation came from /schedule (or user asked for unattended), do not ask interactive browser/tooling questions. Continue only when prior consent is already saved in {user_dir}/CareerNavigator/profile.md under ## LinkedIn:
   • LinkedIn slug: <value>
   • LinkedIn analytics permission: granted
   • LinkedIn analytics mode: chrome | computer | either If any of these are missing, stop with a concise setup message telling the user to run one interactive setup once.

Tool-call guidance

Preferred order: for interactive runs, use the user-approved mode (chrome, computer, or either). If mode is either, try Claude in Chrome first, then computer.

Claude in Chrome path (strict, typed params):

• First try tabs_context_mcp with correctly typed values (for example, createIfEmpty must be the boolean true, not the string "true").
• If context initialization still fails, call tabs_create_mcp, then retry tabs_context_mcp once.
• Then call navigate to the LinkedIn URL.
• If navigate fails because tabId is typed incorrectly, retry once with a numeric tabId (integer, not quoted).
• If still failing, stop with: "Claude in Chrome couldn't initialize a valid tab context. Open Chrome, confirm you're signed into LinkedIn, and rerun linkedin-post-analytics."

No circular fallback loops:

• At most one retry per failing step.
• Do not bounce between tools repeatedly.
• If chosen mode fails, only try the alternate mode when the saved/approved mode is either.

Objective

For your own posts only: open recent activity, collect per-post analytics, append today’s snapshot to tracker.json (networking[]), report a one-line summary per post.

Inputs (ask once if missing)

• Profile path: Read {user_dir}/CareerNavigator/profile.md → ## LinkedIn → LinkedIn slug:. If present, use it. If missing or blank, ask once and save it back to profile.md under that heading before proceeding.
• Consent state (persist): For interactive runs, if missing, ask once and save:
  • LinkedIn analytics permission: granted | denied
  • LinkedIn analytics mode: chrome | computer | either Scheduled runs require permission: granted plus a mode.
• Window: default posts from the last ~60 days (or user’s “several months” if they state it)

Steps

1. Navigate to the recent activity / shares URL above. User should already be logged in; if LinkedIn demands re-auth, stop and notify them.
2. Read-only: do not post, react, comment, or DM. If a post’s analytics control is missing or the panel errors, note and skip that post.
3. For each post in the window: capture post URL (…/feed/update/urn:li:activity:…), date posted (best effort), topic (short label). Open View analytics (bar chart) and read: impressions, members reached, reactions, comments, reposts, saves, sends on LinkedIn, profile viewers from post, followers gained, link visits, plus top audience (industry / seniority / company size) if shown.
4. Read tracker.json. For each post:
   • If networking has type: "linkedin_post" and matching url: append to analytics_history one object: { "date": "YYYY-MM-DD", …metrics… } (same field names as below).
   • Else create a new entry: id = next net-NNN (max existing + 1), type: "linkedin_post", description, url, date_posted, analytics_history: [ { … } ], notes (e.g. auto-discovered + date), outcome: "active".
5. Write tracker.json back. Preserve unrelated keys (applications, etc.).

analytics_history object (use 0 or [] / "—" when unknown):

impressions, members_reached, reactions, comments, reposts, saves, sends, profile_viewers_from_post, followers_gained, link_visits, links (array), top_audience: { "industry", "seniority", "company_size" } (strings).

Report

One line per post: [topic], impressions: X(+Y), reactions: X(+Y), comments: X(+Y), reposts: X(+Y), link visits: X(+Y), [date]. Flag posts with meaningful day-over-day growth (Y) vs the latest prior snapshot. If no posts in the window, say so and exit.

Cadence hint for the user

For users who granted permission and saved mode, morning /schedule runs are supported. Weekly or biweekly cadence is enough for most job-search visibility loops, but daily is acceptable during active networking pushes.