LinkedBoost
託管式LinkedIn MCP伺服器,專為AI助理設計,提供50個已記錄的工具,用於連結席位搜尋、潛在客戶匯入、審核過的推廣序列及收件匣作業。
託管 MCP 伺服器
npx add-mcp 'https://mcp.linkedboost.ai/mcp'可安裝到 Claude Code、Codex、Cursor 等客戶端
文件
The LinkedIn MCP server
A LinkedIn MCP server exposes your connected account as tools your assistant can call. LinkedBoost documents 50 of them: search and import leads, inspect history, prepare sequences, comment on known posts, and work an existing inbox thread. An operator reviews external actions, and the server applies account-level limits. Check completion after a tool queues an action.
What is an MCP server, in one paragraph
Model Context Protocol is an open standard for connecting an AI assistant to an outside system. The assistant discovers a list of typed tools, decides which to call, and passes structured arguments. It is the difference between an assistant that knows about LinkedIn and one that can ask a connected tool for an account-scoped operation. Different servers expose different reads and writes; the tool list and its approval behavior matter more than the MCP label. See the MCP architecture guide for how hosts, clients, and servers fit together.
What your agent can do once connected
| Search & organize | Search through your connected seat, import an authorized CSV, and inspect saved lead groups. |
|---|---|
| Sequences | Prepare a multi-step sequence and review its enrolled people and drafts before external action. |
| Inbox | Read an existing thread, inspect its history, and prepare a reply for review. |
| Pacing | The server documents per-seat invitation caps and warm-up behavior; check your seat's current headroom. |
| Account risk | LinkedIn prohibits unauthorized automation. Approval and caps do not eliminate restriction risk. |
The tool surface, by the numbers
50 typed tools across 7 areas, every one documented with what it does and what you'd say to Claude to use it.
Every tool, by name
The documented tool surface, with an example of what you would say to Claude. A listed tool is not evidence that an action completed on LinkedIn.
Sourcing runs through your own account and authorized CSV imports. There is no bundled contact database. Content tools cover commenting and reacting on other people's posts; publishing your own posts is not part of this build.
What you actually say
You type sentences. The agent picks the tools.
- Import this CSV as a lead group and tell me who I have already spoken to.
- Search my connected account for founders in New York. Show the first results and any filter limitations.
- Here is a post URL and its text. Draft a useful comment from my experience for review; do not publish it.
- Draft a three-step sequence from my seat and show me every draft before anything sends.
- What went out this week, what is queued, and which sequence is underperforming?
- How much headroom does my seat have left today?
Setup in 3 steps
- 1 · Add the connector. In Claude: Settings → Connectors → Add custom connector: or use the one-click link. Full walkthrough on the connect guide.
- 2 · Connect your LinkedIn. One secure flow. Your seat gets a pacing profile and, if it's fresh, a warm-up ramp.
- 3 · Talk to Claude. “Check my account status, then help me research ten prospects. Show the fit evidence and prior contact before drafting. Do not enroll or send.” Review the shortlist before deciding on outreach.
Where the MCP server fits
The hosted endpoint connects your assistant to your LinkedBoost workspace. The account connection, approval decision, server limits, and final action state remain distinct. Other vendors also offer MCP or assistant workflows; the protocol itself does not reduce LinkedIn's terms-of-service risk. See the methodology.
The limits, stated
| Invitation headroom | Per account | Inspect current status before approving outreach |
|---|---|---|
| Warm-up and pacing | Account state | Check the active ramp and queue; capacity can differ by account |
| Messages and engagement | Separate actions | Not an unlimited or restriction-free allowance |
| Sends metered by us | No | Sourcing uses credits; check the current plan terms |
Product caps are operating controls, not a safe allowance from LinkedIn. LinkedIn prohibits unauthorized automation; review the User Agreement and your account's current status before outreach. Limits do not guarantee protection from restriction. Invitation-limit guidance.
Which agents can connect
| Claude | Working connection | Follow the Claude setup guide |
|---|---|---|
| ChatGPT | Working connection | Use the connection options in your workspace |
| Instinct | Working connection | Use the connection options in your workspace |
| Meta Muse | Working connection | Use the connection options in your workspace |
An agency workflow to evaluate first
Use your agency's business-development account to inspect ten prospects or five existing inbox threads. Ask for a relevant shortlist or a draft reply, check the context, then choose one specific action. Confirm the selected account and recipient before approval and the final result afterward.
Account capacity is not a client portal. Native email delivery, CRM sync, client dashboards, and white-label reporting are not offered. Company overlap needs operator review; history tools do not guarantee that every outside interaction is known.
Agency workflow guide Compare 1, 3, and 10-account plans
Questions
Is LinkedBoost an official LinkedIn MCP server?
No. LinkedBoost is independent of LinkedIn. It connects an account to an assistant through hosted MCP tools; the protocol does not imply LinkedIn approval or endorsement.
Which agents work with it?
Claude, ChatGPT, Instinct, and Meta Muse connections are confirmed working. This site includes a Claude setup walkthrough; connection settings and controls differ by assistant.
What's the MCP server URL?
The hosted endpoint is https://mcp.linkedboost.ai/mcp. Sign in through the connector to use your workspace.
Can it get my account restricted?
Yes. LinkedIn prohibits unauthorized automation, and limits or review do not remove account risk. Our restriction rate is still measuring; see /safety for the proposed methodology.
How is this different from a LinkedIn scraper?
It exposes account-scoped write tools as well as search and read tools, including invitations, messages, replies, and comments. A tool listing or queued action is not proof that the action completed.
Do I need Sales Navigator?
It is not required to connect. Search and InMail access depend on the entitlement of your connected account. Check the available filters and account status before planning a workflow.
Where do the leads come from?
Your connected LinkedIn seat and CSV data you are authorized to import. There is no bundled contact database or native CRM sync. Check imported history and suppression state before outreach.
Can it post content for me?
It can comment on and react to other people's posts, which is the engagement half of content. Publishing your own posts is a separate capability and is not in this build.
When does a sequence start outreach?
Creating a sequence prepares a plan. Enrollment can queue its scheduled actions, so review exact recipients, sender, copy, and cadence before approving it. Inspect the queue and final action status; do not infer completion from a draft.
What happens when I hit an account limit?
Check account status and the returned action result. Product caps can prevent an action from proceeding. Do not repeatedly retry a capped or queued request; inspect its state and available headroom first.
Go deeper
[
What a LinkedIn AI agent can actually run for you
The capability map, tool by tool, and where the agent is deliberately not in charge.
](https://linkedboost.ai/blog/linkedin-ai-agent)[
What actually gets a LinkedIn account restricted
The behavioral signals that carry weight, and what each restriction tier means.