How do you set up a LinkedIn MCP server with Claude?
Add https://mcp.linkedboost.ai/mcp as a custom connector in Claude, complete the Google sign in, then tell the assistant to connect your LinkedIn account. Verify by asking what is in your workspace and expecting real numbers back. A connector showing as added is not proof that any of it works.
The four steps
One: add the custom connector. The URL is https://mcp.linkedboost.ai/mcp. In claude.ai or Claude Desktop the path is Settings, then Connectors, then Add custom connector, then paste that URL.
Two: complete the Google sign in when the browser window opens. That authenticates you to the workspace, which is where your lead groups, sequences and history live. It does not touch LinkedIn, and it will not ask for your LinkedIn password at any point, here or later.
Three: in a chat, say "connect my LinkedIn account". The assistant calls the connect tool and hands back the link that binds your seat. Four: verify, which is the step almost everyone skips and the only one that proves anything.
Three separate states, and the tick only covers one
Transport connected means Claude can reach the server. Workspace authenticated means the server knows who you are. LinkedIn linked means there is an account behind the tools that can actually do something. The connector appearing in your settings list is the first of those, on its own.
This is the most common support pattern we see. Someone adds the connector, sees the tick, asks for a message to go out, and gets a confident-sounding answer about something that did not happen, because no account was ever linked. The interface was telling the truth about the only state it knows.
The reason this matters beyond setup is that the third state can fail later, on its own, without disturbing the first two. LinkedIn invalidates sessions for ordinary reasons, and when it does the connector still shows as added and the sign in is still valid. Only the account link is gone. The same verification question is the diagnosis for that too.
The verification that actually works
Ask: "what is in my workspace?" A working connection answers with numbers. How many lead groups. How many sequences. How many conversations. Which LinkedIn account is linked, and whether the link is currently healthy.
A broken connection answers in prose: general, plausible, no figures, often an offer to help you get started. If there are no numbers, nothing is connected, whatever the settings page shows. Numbers are the test because numbers can only come from a tool call that returned.
Two follow-ups are worth asking in the same breath. Ask which LinkedIn account is linked and check that it is the one you meant, because people with two profiles link the wrong one more often than you would expect. Then ask how much outreach volume is left today, which confirms the caps are visible to the assistant rather than something it will discover by being refused mid-sequence.
What the first hour should look like
Get people in first. Search through your own LinkedIn seat, or import a CSV you already own. Those are the two sources and there is no third, because there is no bundled contact database behind either of them. An import earns its keep twice on day one, because your contacted history becomes suppression memory immediately and nobody already on it can receive a duplicate first touch.
Then create a sequence, which sends nothing by itself. Enroll a lead group, which queues drafts rather than sending them. Read the drafts, approve the ones you would have sent yourself, and watch the first replies stop their own follow-ups.
Expect the caps to hold you back at the start, deliberately. A new seat runs at 20 invitations a day and 80 a week, and the warm-up ramp multiplies that by 0.5x in week zero, 0.75x in week one and 1.0x from week two. These are enforced server-side. There is no settings field to raise them, and that is the feature.
Where setup usually goes wrong
An expired LinkedIn session. Password changes, security challenges and signing out on another device all break the link, and the tool error that follows can read like a product fault. Ask the assistant to check the connection, then reconnect the account.
The wrong expectation about clients. Claude is first-class today across claude.ai, Desktop and Claude Code, and any MCP-capable client can connect to an open protocol. ChatGPT and Cursor are on our roadmap rather than shipped, so plan around Claude for now.
Expecting something that is not in this build. There are no CRM, Zapier, HubSpot, Salesforce or webhook tools, so nothing appears in another system after a send. And the assistant will not publish posts as you. Commenting and reacting on other people's posts is the whole of the content surface here.
The checklist worth running before your first send
Numbers came back when you asked what is in your workspace. The linked account is the one you intended. A lead group exists with people in it, from a search through your seat or from an import. A sequence exists and has sent nothing, because creating one never does.
Drafts are queued and you have read them rather than skimmed them. You know how much volume is left today. And you can say out loud why each person is on the list, because that is the variable that moves results furthest: across two of our audiences with comparable copy, acceptance was 32% and 2.6%.
If all of that is true, the first approval is a real decision rather than a formality, which is the state the product is built around. Across 861 conversations over three years on five seats, our own acceptance averaged 41%, and nothing in that history left without somebody reading it first.
The three states, and how to tell them apart
| State | What it means | How you check it |
|---|---|---|
| Transport connected | Claude can reach the server | The connector appears in Settings |
| Workspace authenticated | The server knows who you are | Google sign in completed without an error |
| LinkedIn linked | There is an account behind the tools | Ask what is in your workspace and get numbers |
Questions people ask next
How do I add LinkedBoost to Claude?
Settings, then Connectors, then Add custom connector, then paste https://mcp.linkedboost.ai/mcp. Complete the Google sign in, then say "connect my LinkedIn account" in a chat.
How do I know the LinkedIn MCP server is actually working?
Ask what is in your workspace. A real connection answers with counts of lead groups, sequences and conversations. Prose with no numbers means nothing is connected yet.
Does adding the connector connect my LinkedIn account?
No. Adding the connector and signing in authenticates you to the workspace. Linking the LinkedIn account is a separate step, started by asking the assistant to connect it.
Can I use this with ChatGPT or Cursor?
Claude is first-class today, and the protocol is open, so any MCP-capable client can connect. ChatGPT and Cursor are on our roadmap rather than shipped.
Why this page exists: LinkedBoost connector onboarding flow, documented September 2026
LinkedBoost is the LinkedIn MCP server: your agent sources, drafts, sends and works the inbox, inside caps the server enforces rather than suggests.
Related answers
Search your own seat, import your own CSVs, research a person into a persistent column, run approval-gated sequences, send invitations and messages, comment and react on other people's posts, work the inbox, and report on all of it. It will not publish your posts and it does not sync anywhere.
An MCP server gives an AI assistant hands. Without one the assistant can describe what to do; with one it can do it, in your account, against real state. The value shows up only when the task is mechanical, repeated and needs live data.