LinkedIn MCP server: what it is, how to choose one, how to set it up
A LinkedIn MCP server lets AI assistants read and act on LinkedIn. The tools, hosted vs self-hosted, a checklist, and setup for Claude Code, Cursor and more.
Finchreach teamMcpLinkedinClaude codeCursorHow to
The short answer
A LinkedIn MCP server is a service that gives an AI assistant LinkedIn “tools”: read your feed, search people, draft a post, reply to a message. MCP (Model Context Protocol) is a standard plug that lets AI apps use outside tools, so one server works with Claude, Cursor, Codex, n8n and any other client that speaks MCP.
Finchreach is a hosted LinkedIn MCP server at https://mcp.finchreach.com/mcp. You connect your LinkedIn account in a dashboard, create an API key, and add one line of config to your assistant. Reads happen right away. Every write (post, comment, reaction, message, invitation) waits in an approval queue until you approve it.
If you only want the concept, our shorter explainer what is a LinkedIn MCP server covers it. This post is the practical version: what the tools look like, how the options differ, and how to set one up.
How does a LinkedIn MCP server work?
Three pieces are involved:
- Your assistant (the MCP client), such as Claude Code or Cursor. It reads the server’s list of tools and their descriptions.
- The MCP server. It turns a tool call such as “create a post with this text” into an action on LinkedIn, and returns the result as structured data.
- Your LinkedIn account, connected to the server in some way: an official API grant, or your own logged-in session.
When you ask “what did people say about my post yesterday?”, the assistant picks the notifications tool, reads the result, and answers in plain language. When you ask it to reply, it picks the comment tool, and it can react to a post too; both are live and wait for your approval. You never write the API calls yourself.
Finchreach is a remote server using MCP’s Streamable HTTP transport. Your assistant connects over HTTPS and authenticates in one of two ways: OAuth sign-in, where you sign in to Finchreach and approve on a consent page, or an API key in a header (Authorization: Bearer lk_... or X-API-Key). Nothing runs on your machine unless your client only runs local servers and needs a small bridge.
What tools does a LinkedIn MCP server give my assistant?
Here is Finchreach’s full tool list as of October 2026. Other servers vary, but this is a useful baseline for comparing them.
| Tool | What it does | Approval |
|---|---|---|
linkedin_list_accounts | Your connected accounts and their status | n/a |
linkedin_get_profile | Your own profile: name, headline, public id | n/a |
linkedin_get_feed | Posts from your home feed, paginated | n/a |
linkedin_get_notifications | Latest notifications and unseen counts | n/a |
linkedin_get_conversations / linkedin_get_messages | Your inbox and the messages in one conversation | n/a |
linkedin_search | Search all, posts, people (by network degree) or companies | n/a |
linkedin_create_post | Publish a text post as you | Required |
linkedin_comment_on_post / linkedin_delete_comment | Comment on a post, or remove your comment | Required |
linkedin_react_to_post / linkedin_unreact | Add or remove a reaction | Required |
linkedin_send_dm | Message an existing conversation or new recipients | Required |
linkedin_send_invitation | Connection request with an optional note (up to 300 characters) | Required |
linkedin_get_received_invitations / linkedin_accept_invitation | Review and accept invitations you’ve received (coming soon, not live yet) | Coming soon |
linkedin_get_action | Status of a queued write: pending, approved, rejected, executed, failed, expired | n/a |
What’s not there matters too: no image or video posts, no scheduling, no company pages and no analytics. A server that clearly says what it can’t do is easier to trust than one that promises everything.
Should I use a hosted or a self-hosted LinkedIn MCP server?
You’ll find open-source LinkedIn MCP servers you run yourself, developer APIs you build on, and hosted servers like Finchreach. The trade-offs:
| Self-hosted open source | Hosted (e.g. Finchreach) | |
|---|---|---|
| Setup | Install, configure, keep it running | Sign up, connect, paste a URL and key |
| Where your session lives | On your machine or server | With the provider, so check how it’s stored |
| Network location | Your own IP, which may change | Depends on the provider |
| Human approval | Usually none, or only your client’s prompt | Built in, if the provider offers it |
| Maintenance when LinkedIn changes | You, or the maintainer | The provider |
| Cost | Your time and hosting | The provider’s price |
Neither is automatically safer. Self-hosting keeps your session on your hardware but leaves the approval logic, updates and security to you. Hosting moves that work to a provider, so you need to trust how they store your session and keys.
Be clear on how any server reaches LinkedIn. LinkedIn’s official APIs are limited for personal accounts: messaging and most reading aren’t open to general developers. So most tools that read your feed or send messages, Finchreach included, act through your own LinkedIn session. In Finchreach, you paste session cookies in the dashboard, and each account’s traffic goes through one stable proxy location instead of jumping around. That’s practical, but it’s also why the LinkedIn User Agreement applies (more below).
How do I choose a LinkedIn MCP server?
Before you connect any server to your real account, check:
- Does every write wait for a person, on every client? A confirm prompt in your assistant is good; a server-side queue that the assistant can’t skip is better.
- Can you scope keys? A key that can only read is ideal for research. A key that can submit but not approve keeps the decision with you.
- Can you revoke a key instantly and see what each key did?
- Does it say clearly what it can’t do? Vague claims about “full LinkedIn automation” are a warning sign.
- Does it talk honestly about LinkedIn’s rules? Any provider promising “ban-proof” automation is guessing at best.
- Which transport and auth does it use, and does your client support them? A header-only server won’t work in a client that only offers OAuth, such as the ChatGPT app.
If you’re comparing Finchreach with content-scheduling tools, see Finchreach vs Taplio. Those tools are built around a web editor and a calendar. An MCP server is built around your assistant.
How do I set up the Finchreach LinkedIn MCP server?
- Create an account at finchreach.com/sign-up.
- Connect your LinkedIn account under Accounts.
- Create an API key under API keys and store it in an environment variable, for example
FINCHREACH_API_KEY. - Add the server to your client (examples below, more in the docs).
- Ask your assistant to list your LinkedIn accounts to test the connection.
Claude Code (guide):
claude mcp add --transport http finchreach https://mcp.finchreach.com/mcp \
--header "Authorization: Bearer $FINCHREACH_API_KEY"Cursor, in .cursor/mcp.json or ~/.cursor/mcp.json (guide):
{
"mcpServers": {
"finchreach": {
"url": "https://mcp.finchreach.com/mcp",
"headers": { "Authorization": "Bearer ${env:FINCHREACH_API_KEY}" }
}
}
}VS Code with GitHub Copilot uses servers instead of mcpServers in .vscode/mcp.json (guide). Windsurf, Cline, Codex CLI, Gemini CLI and n8n all accept a header too; see the guides for Windsurf, Codex and n8n.
For clients that only run local servers, the mcp-remote bridge adds the header for you:
npx -y mcp-remote https://mcp.finchreach.com/mcp --header "Authorization: Bearer ${FINCHREACH_API_KEY}"Chat apps that can’t send a static key connect with OAuth instead. As of October 2026, ChatGPT (developer mode), claude.ai, Claude Desktop and the Gemini app add Finchreach as a custom connector from the URL; you sign in to Finchreach and approve. See how to let ChatGPT post on LinkedIn and the Claude setup.
What should I ask a LinkedIn MCP server first?
Start with reads, so you can see what the assistant sees before it writes anything:
Ask your assistant“Using Finchreach, show my LinkedIn profile headline and summarize my latest notifications in five bullets.”
Ask your assistant“Search LinkedIn for people in my 2nd-degree network who work on developer relations at database companies. List ten with their headlines. Don’t contact anyone.”
Then try one write and follow it through the queue:
Ask your assistant“Draft a two-sentence comment on the most recent post in my feed about remote work, submit it, and tell me the action id so I can approve it.”
Is using a LinkedIn MCP server against LinkedIn’s rules?
It can be. As of October 2026, LinkedIn’s User Agreement (section 8.2) says members must not use “bots or other unauthorized automated methods” to access the service, send messages, or create, comment on, like or share posts. Whether a given tool falls under that is LinkedIn’s call, not ours.
Automating a personal LinkedIn account carries risk under LinkedIn’s User Agreement. Finchreach’s approval queue and user-paced actions lower that risk; they don’t remove it.
What we’ve built around that: a person approves each write, nothing is sent in bulk, and keys can be read-only. Our full take is in LinkedIn automation without getting banned and how to automate LinkedIn with AI safely.
LinkedIn MCP server FAQ
Is Finchreach an official LinkedIn product? No. It’s an independent service and isn’t affiliated with LinkedIn or Microsoft.
Can I connect more than one LinkedIn account? linkedin_list_accounts returns every account connected to your workspace, and tools take an account id. Each account is still a real person’s account, under that person’s User Agreement.
Does the assistant see my password? No. The assistant only sees the Finchreach API key you give it and the tool results. You connect LinkedIn in the Finchreach dashboard, not in the chat.
What if my LinkedIn session expires? The account shows as expired or checkpoint in linkedin_list_accounts. Reconnect it under Accounts.
Where do I go next? The quickstart and tool reference, or the Claude Code integration.