Insaight — LinkedIn Intelligence Tool

Scrapes LinkedIn company pages and personal profiles via Apify, stores in SQLite, and exposes them to Claude via MCP for prospect research and cold outreach.

Skills live in skills/<name>/SKILL.md — each is independently invocable and they chain conversationally (research → outreach → save).

Setup

User config lives in ~/.insaight/config.md (created by get_config() on first call) — not in this file. The block below is the template; skills call insaight:get_config to read it.

NOTION_WORKSPACE:     YourWorkspaceName   # your Notion workspace / team name
NOTION_RESEARCH_PAGE: Prospect Research   # parent page for research briefs
NOTION_OUTREACH_LOG:  Sent Log            # your sent-messages archive (READ-ONLY)
COMPANY_NAME:         YourCompany         # your company (used by draft-post)
COMPANY_LINKEDIN:     your-company-slug   # your LinkedIn company page slug

Insaight MCP — Tool Reference

All tools must be loaded first via tool_search(query="insaight").

Efficient reading pattern

1. list_accounts()                    → find the correct slug
   └─ not found? → scrape_profile(url) to fetch and store fresh posts

2. list_posts(account=slug, limit=50) → survey what exists (slim, token-cheap)

3. get_posts(urns=[...])              → read full content for chosen posts only
   OR search_posts(query=...)         → targeted lookup when you know what to look for

4. list_people(account=slug)          → check stored leadership
   └─ empty? → scrape_people(url, job_titles=["CEO","Founder","CTO","Head of"])

5. scrape_person_profile(url=...)     → enrich ONE person with full profile
   (experience, education, skills, volunteer, languages, projects, recommendations)
   Use for commonality mining — overlapping companies, schools, volunteer work.

6. list_comments(post_url=...)        → read commenters on a post (free if already scraped)
   └─ empty? → scrape_post_comments(post_url, include_replies=True)
   Use for mining a thread for warm leads, decision-makers, competitor mentions.

Never call get_posts on all URNs at once. Select the most signal-rich posts to stay within the 20-URN limit per call.

Cost warnings

scrape_profile and scrape_people call Apify (costs tokens/money). Only use them when the account isn't tracked yet or data is stale. list_people is free/instant — always try it first.

Account slug format

• Company page: acme-charging (matches the LinkedIn URL slug)
• Personal profile: jane-doe-12345678 (matches the LinkedIn profile ID)

Engagement benchmarks (example: EV charging niche, NL/EU B2B — adjust for yours)

• < 10 likes = low
• 10–40 likes = normal
• 40+ likes = high signal post, read in full

Outreach memory loop

The ledger + memory system replaces re-reading a raw sent-log every session:

1. Log every send (log_outreach) with the exact sent text + variant + hook_type.
2. Record every outcome (record_outcome) when the user reports a reply, meeting, or gives up (ghosted). The user decides when something is ghosted.
3. After REFLECT_EVERY outcomes (env var, default 10), record_outcome returns reflection_due: true → offer the insaight-reflect skill.
4. Reflection proposes updates to ~/.insaight/memory/style.md (voice) and ~/.insaight/memory/playbook.md (strategies with evidence counts) — the user approves before update_memory is called. Never overwrite memory silently.
5. draft-outreach reads get_memory(), not raw history — constant token cost regardless of ledger size.

Playbook discipline: every claim carries its evidence ("question hooks: 4/9 replied"); below n=10 a pattern is a hypothesis, not a rule.

Tips

• Personal profiles vs company pages: Personal profiles (founders, CEOs) often reveal more strategic signal than corporate pages. If both are tracked, read both.
• Non-English content: Many European prospects post in their own language. Translate mentally and include in your analysis — don't skip non-English posts.
• Low post volume: If the account has < 10 posts, supplement with web_search("[company name] EV charging news 2025 2026").
• Category field: The category field in Insaight data is currently unpopulated. Do your own thematic grouping from snippets.
• Time gaps: A long gap in posting (> 3 months) is itself a signal — note it.

Notion Integration

Two Notion locations matter for every skill run:

1. [NOTION_RESEARCH_PAGE] (read + write)

Parent page containing one sub-page per researched person/company.

• Location: [NOTION_WORKSPACE] → [NOTION_RESEARCH_PAGE]
• Page naming: [Name] — [YYYY-MM-DD]
• Used by: research skills (read existing page as context, then write updated brief), save-notion (the writer)
• Find it:

  notion:search(query="[NOTION_RESEARCH_PAGE]")

2. [NOTION_OUTREACH_LOG] (READ-ONLY — legacy fallback)

The outreach ledger + memory system (see "Outreach memory loop") has replaced this page. Sends live in the SQLite outreach table; style lives in ~/.insaight/memory/style.md. Only fall back to this Notion page when get_memory() reports nothing learned AND list_outreach() is empty (fresh install with pre-existing Notion history).

• Location: [NOTION_WORKSPACE] → [NOTION_RESEARCH_PAGE] → [NOTION_OUTREACH_LOG]
• NEVER write to this page. The user maintains it manually.
• Find it:

  notion:search(query="[NOTION_OUTREACH_LOG]")

Workflow integration

Session caching

Fetch memory (get_memory) once per conversation and reuse it. Same for any legacy Notion pages you had to load — never re-fetch on iteration.

Notion MCP loading

The Notion MCP is separate from insaight. If notion:search / notion:fetch / notion:create-pages aren't available, load them first:

tool_search(query="notion search fetch create")

If Notion MCP isn't connected at all, tell the user:

"Notion MCP isn't connected. Connect it at claude.ai → Settings → Integrations, then retry. Research will proceed but won't auto-save."