LinkedIn MCP Server

MCP server for LinkedIn. It lets AI assistants like Claude scrape profiles and posts, build a voice profile from someone's posts, and write or improve posts in that voice. Scraping and AI work are delegated to dedicated LinkedIn actors (profile scraper, post scraper, voice trainer).

BYOK (Bring Your Own Key): real data needs your own LinkedIn cookies, and the AI tools need your own Claude or OpenAI API key.

---

How it runs

• On Apify: each run executes one tool (the tool input) with toolArguments, writes one result row to the dataset, and exits.
• Locally (Claude Desktop or any MCP client): node dist/main.js starts an MCP server over stdio. Credentials can come from tool arguments or the LINKEDIN_COOKIES, ANTHROPIC_API_KEY and OPENAI_API_KEY environment variables.

Live mode never returns sample data. If cookies or keys are missing, or a scrape fails, you get a result row with success: false, an error message and an errorType, and nothing is charged for that tool call.

---

Quick start

See the available tools (free)

{
  "tool": "list_tools"
}

Scrape a profile

{
  "tool": "scrape_profile",
  "toolArguments": { "profileUrl": "https://www.linkedin.com/in/someone/", "maxPosts": 20 },
  "linkedinCookies": "[{\"name\":\"li_at\",\"value\":\"...\"}]"
}

Build a voice profile and write a post

1. Run analyze_voice with toolArguments {"profileUrl": "..."} (needs cookies) or {"posts": [...]}, plus claudeApiKey or openaiApiKey.
2. Run generate_post with {"topic": "...", "voiceProfile": <the voiceProfile from step 1>} and the same AI key.

Sample output

Set demoMode to true to get labeled sample output (isSample: true) for any tool. Sample output is never charged.

---

Tools

Tool	Needs	Notes
list_tools	nothing	Tool catalog with input schemas
scrape_profile	cookies	Profile details and recent posts
scrape_posts	cookies	Posts by profileUrl, hashtag or keyword
analyze_voice	AI key (+ cookies with profileUrl)	Returns a voiceProfile object
generate_post	AI key + voiceProfile	Post in that voice
improve_draft	AI key + voiceProfile	Improved draft
suggest_topics	AI key + voiceProfile	Topic ideas
analyze_performance	posts, or profileUrl + cookies	Engagement summary and recommendations
search_people		Not supported yet (returns an error)
get_voice_profile, list_voice_profiles		Not supported yet: voice profiles are not stored between runs. Pass the voiceProfile object instead

---

Input parameters

Parameter	Type	Default	Description
tool	string	list_tools	Tool to run once on Apify
toolArguments	object	{}	Arguments for that tool
linkedinCookies	string		JSON array of LinkedIn cookies (must include li_at)
linkedinEmail / linkedinPassword	string		Alternative to cookies, passed to the scraper actors
claudeApiKey / openaiApiKey	string		AI key for the voice and writing tools
demoMode	boolean	false	Labeled sample output, never charged
webhookUrl	string		Receives the result row via POST
mode	string	stdio	Legacy. demo behaves like Demo Mode

---

Pricing

Pay per event:

Event	When it is charged	Price
Profile Searched	Once per tool call that returned real LinkedIn profile or post data: scrape_profile, scrape_posts, and analyze_voice or analyze_performance when they scrape a profileUrl	$0.05

Apify also applies its standard actor start and per-result platform events listed on the pricing tab. Sample output, errors, list_tools and the AI-only tools do not charge Profile Searched.

The delegated actors (LinkedIn Profile Scraper, LinkedIn Post Scraper, LinkedIn Voice Trainer) run in your Apify account and bill their own events.

---

Troubleshooting

• MISSING_LINKEDIN_COOKIES: add linkedinCookies, or set demoMode to true for sample output.
• MISSING_AI_KEY: add claudeApiKey or openaiApiKey.
• DELEGATE_RETURNED_SAMPLE or NO_RESULTS: the scraper could not log in or found nothing. Export fresh cookies and try again.
• DELEGATE_TIMEOUT: the delegated scrape took longer than 5 minutes and was stopped. Try fewer posts.

---

Built by John Rippy | Actor Arsenal