MCP Linkedin

Skill: LinkedIn Publishing via MCP

Important: Agent Session Limitation

MCP tools (including all linkedin_* tools) are not available in --agent sessions (e.g. when launched via claude --agent <name>). They only load in top-level claude sessions.

If you are inside an --agent session, you must spawn a sub-agent to use these tools:

# Write your prompt first, then pipe it:
cat /tmp/linkedin-prompt.md | claude -p --dangerously-skip-permissions

The sub-agent runs as a top-level claude session and will have full access to all LinkedIn tools.

Unipile Account Required

All 3 tools use Unipile as the LinkedIn API provider. You need a Unipile account with your LinkedIn account connected. Get your UNIPILE_API_KEY and UNIPILE_DSN from the Unipile dashboard and add them to your MCP config.

Available Tools

Safe Publishing Workflow

CRITICAL: Always use dry_run first

linkedin_publish and linkedin_comment both default to dry_run: true. Never set dry_run: false without explicit user confirmation.

Tool: linkedin_publish

Step-by-step workflow

1. Call with dry_run: true (default — can omit the parameter):

   linkedin_publish({ text: "Your post text here", dry_run: true })
   

2. Present the preview to the user. Include:

   • The final post text (with mention placeholders shown as {{0}}, {{1}})
   • Character count vs 3000 limit
   • Media validation results (valid/invalid per file)
   • Resolved mentions (which company names were found vs not found)
   • Any warnings (e.g. "character limit exceeded", "1 media item could not be processed")
   • ready_to_publish flag

3. Ask the user for confirmation before publishing. Example: "The post looks good (142 chars, 1 image attached). Publish it?"

4. Only after confirmation, call with dry_run: false:

   linkedin_publish({ text: "Your post text here", dry_run: false })
   

5. Report back the post URL. Construct it from the post_id:

   https://www.linkedin.com/feed/update/urn:li:activity:{post_id}/
   

   This step is mandatory — always include the clickable URL in your response.

6. The publish response includes auto_like — the post is automatically liked after publishing.

Post formatting best practices

• Hashtags: Use 3-5 relevant hashtags at the end, preceded by a blank line
• Line breaks: Use double newlines for paragraph breaks in LinkedIn posts
• Character limit: 3000 characters max. Keep it under 1300 for better reach
• Emoji: Use sparingly at the start of bullet points or sections
• Mentions: Pass company names in the mentions array — they are resolved automatically

Tool: linkedin_comment

Step-by-step workflow

1. Call with dry_run: true:

   linkedin_comment({
     post_url: "https://linkedin.com/feed/update/urn:li:activity:12345",
     text: "Great insight!",
     dry_run: true
   })
   

2. Show the preview (post URN, comment text, character count).

3. Ask for confirmation.

4. Call with dry_run: false after confirmation.

Tool: linkedin_react

linkedin_react({
  post_url: "https://www.linkedin.com/feed/update/urn:li:activity:7123456789012345678/",
  reaction_type: "celebrate"
})

Note: linkedin_react has no dry_run — confirm with user before calling.

Reaction types: like, celebrate, support, love, insightful, funny.

Error Handling

Auto-Like Behavior

linkedin_publish automatically likes the post immediately after publishing. This is reported in the response as:

{ "auto_like": "liked" }

If the auto-like fails (e.g. rate limit), the response contains an error message but the post itself is still published successfully.

Warnings

• Never auto-publish — always show the dry_run preview and wait for explicit user confirmation.
• Media files downloaded from URLs are stored in /tmp/mcp-linkedin-media/ and cleaned up after publish.
• Mention resolution is best-effort. Unresolved mentions appear in warnings — the post can still be published without them.