Back to guides
/Linked API

Claude for Sales: Add LinkedIn to Call Prep and Follow-Up

Most "Claude for sales" guides start in the CRM. This one is about one call, and the part of preparing for it that lives on LinkedIn: who this person is now, who you already know at their company, and what you last said to each other.

The short version. Claude is already good at the CRM half of a sales call: Anthropic's Sales plugin researches the account, preps the agenda and summarizes the call from your CRM, call recorder, data providers and public web search. What it does not have is your LinkedIn account – who at that company you are already connected to, what you two last said in your LinkedIn thread, and a way to reply there. Connect Claude to your own LinkedIn account and it can read the person's recent posts, your connections at the account and your thread before the call, then draft the LinkedIn follow-up after it: a reply in that thread, or a connection note for someone new on the call. You approve every message before it is sent.

Here is one discovery call prepared from a single ask – the setup is below:

Your Agent
runs on your own LinkedIn account
Discovery call with Acme on Thursday: Jane Doe, VP Sales, and Raj Patel, RevOps lead; I've pasted their LinkedIn URLs and Acme's company page. Using my LinkedIn account, read each attendee's profile, experience, and posts and comments from the last 90 days, check whether we're connected, load our message history and read our thread, and find who I know at Acme in sales roles. Write a one-page brief with three openers, each tied to a post or a message, and the questions you couldn't answer. Don't send anything.
Agentvia MCP Server
Used Linked API: fetch_person, check_connection_status, fetch_company, sync_conversation, get_inbox
  • profile, experience, and posts and comments from the last 90 days
  • connected to Raj; not connected to Jane, no invitation pending
  • employees with "sales" in their position, at 1st or 2nd degree to you
  • your message history with each attendee, loaded once
  • the thread with each attendee, by their profile URL

Brief ready for both attendees. You're connected to Raj and have a thread with him; Jane is a 2nd-degree connection with no message history. At Acme, one account executive is a 1st-degree connection and four people in sales roles are 2nd-degree – people to look into, not a ready introduction path. Open question: the size of Jane's previous company isn't on her profile. Nothing has been sent.

WhoRole and tenureRecent activityYou and themOpener
Jane DoeVP Sales, 2 years 4 months; before that, sales director at a payments companyTwo posts this month on ramping new reps faster; a comment on a thread about forecast accuracy2nd degree, no message historyHer ramp-time post; the forecast-accuracy thread
Raj PatelRevOps lead, 8 monthsNothing in the last 90 days1st degree; in June you sent him a webinar link and he replied "will pass to the team"His June reply about passing it to the team

nothing is sent from your account – the follow-up after the call waits for your approval

Works withClaudeCodexChatGPTCursorOpenClaw
Example run – the names and details are invented for illustration. Claude reads what your account can see: a private or limited profile stays limited. Works in Claude, ChatGPT, Cursor or Claude Code through the MCP server or the CLI.

What Claude already does for sales – and what your LinkedIn account adds

Anthropic's Claude for Sales offer is a Sales plugin with "ready-made skills for account research, call prep, call summaries, competitive intelligence, pipeline review", and connectors to Salesforce, HubSpot, Gong, Fireflies, Clay, ZoomInfo, Apollo, Outreach, Docusign, email and calendar. Its own Cowork tutorial frames the job as "research any account before the first call", then run /call-summary afterwards. The plugin's account research works "via enrichment (web search + any connected data providers)", and its call prep describes attendees from their CRM title.

That covers a lot. What none of the connectors in those sales materials do is link your own LinkedIn account, so the parts of a call that only your account knows stay outside it:

MomentWhat Claude uses todayWhat your LinkedIn account adds
Before the call – the accountCRM records, data providers, web searchWho works there and at what connection degree to you; the company's recent posts
Before the call – the personCRM title, public web searchTheir profile and experience, recent posts and comments as your account sees them, and whether you are connected
Before the call – your historyEmail threads, call recordingsYour LinkedIn message thread with them
After the callAn email draft and a CRM updateA reply in the LinkedIn thread, a connection note for a new attendee, a check for their reply

Claude can search the public web, but it has no access to your LinkedIn account by itself. The account reads and the sends below come from a tool connected to your account.

Give Claude access to your LinkedIn account

  • claude.ai, the Claude Desktop app or Cowork: add Linked API as a custom connector. One connector covers all three (step-by-step).
  • Claude Code: use the same MCP server, the agent-friendly CLI or the ready-made skill (npx @linkedapi/skills, see the LinkedIn skill).
  • What it runs on: a dedicated cloud browser on your own account, at a human pace, inside the per-action limits and working hours you set. Linked API never stores your password.
  • Turn on inbox monitoring once. Ask Claude to run sync_inbox once for the account. It makes new replies show up when Claude checks the inbox, and it is what fires inbox webhooks if you automate the reply check later. Loading one person's history (sync_conversation) adds only that person's new messages, for a limited period (30 days by default), and fires no webhooks.
  • Sales Navigator: the Sales Navigator tools (nv_) need a Sales Navigator subscription on the account and a Linked API Plus seat.

The full tool list is in available tools, and the options for giving an agent access to your LinkedIn account – what each connector holds and where actions run – are compared in the AI-agent guide.

Before the call: the brief Claude builds from LinkedIn

Give Claude the meeting, the attendees' LinkedIn URLs and the company page, and tell it what to read – the run at the top of this page is one such ask. The tools behind each step:

StepToolWhat it returns
Person, experience and recent activityfetch_person with experience, posts and comments since a dateProfile, roles, recent posts and comments
Connection statuscheck_connection_statusConnected, not connected, or an invitation pending either way
Who you know at the accountfetch_company with employees filtered by position and connection degree (1st, 2nd, 3rd+)Matching employees with name, headline and profile URL
Your threadsync_conversation once per person, then get_inbox with the person's URLThe whole known conversation with that person

Two boundaries to keep in the brief. The 2nd-degree filter returns employees you are two steps from, not the mutual connection who could introduce you – those are people to investigate, not a ready path. And Claude reads what your account can see: a private or limited profile stays limited.

After the call: the LinkedIn follow-up

Paste your call notes – from /call-summary or your own – and ask for the LinkedIn half:

Your Agent
runs on your own LinkedIn account
Here are my notes from the Acme call. Using my LinkedIn account, draft a reply to Raj in our existing LinkedIn thread that confirms what we agreed and the next step, and a connection note to Jane of 200 characters or less that references the call. Show me both drafts and wait – never send without my explicit approval.
Agentvia MCP Server
Used Linked API: get_inbox, check_connection_status
  • your thread with Raj, loaded before the call
  • Jane – still not connected, no invitation pending

Two drafts ready, written from your notes. Nothing has been sent.

ToWhere it goesDraft
Raj PatelYour existing LinkedIn thread"Thanks for today, Raj. As agreed, I'll send the pilot scope by Friday, and we'll go through it with Jane next Thursday."
Jane DoeConnection request with a note (109 characters)"Hi Jane – thanks for your time today. Glad to connect and to follow up on getting new reps productive faster."

nothing goes out until you approve each draft; then Claude sends both at a human pace on your account

Works withClaudeCodexChatGPTCursorOpenClaw
Example run – the drafts draw only on your notes and what Claude read before the call. Works in Claude, ChatGPT, Cursor or Claude Code through the MCP server or the CLI.

When you approve, Claude sends them: the reply with send_message into the existing thread – by its thread ID when the conversation has one, or by the person's URL when it does not (history loaded from a profile can carry no thread ID) – and the invitation with send_connection_request and your note. Connection notes are limited on free LinkedIn accounts, which get only a few personalized invitations a month; the connection request guide covers the details.

A day later, ask "did Raj or Jane reply?" and Claude reads get_inbox for each attendee's URL. Two things to know about that answer: the inbox updates on a periodic sync, so a message from minutes ago may not be there yet, and an empty result means no reply has been captured yet – not that none was sent. Replies are captured once inbox monitoring is on – and for a person whose history Claude loaded, during that load's syncing period (30 days by default).

Route replies to the deal owner in your own tools

Developers can let replies come to them instead of asking. With inbox monitoring turned on once per account (syncInbox), every new message reaches your endpoint as an inbox.messageReceived webhook; without that step, it receives nothing. Key the call attendees by the hashedUrl from their fetchPerson result – the form the inbox reports on every message, while the public URL can be missing – and route a matching reply to the deal owner in Slack or your CRM. Deliveries are at least once, so deduplicate on the event id. The same calls run through the REST API, the Node and Python SDKs – parseWebhookEvent / parse_webhook_event read the payload – and the shell CLI; the docs cover each action. Whole-inbox automation – every thread, not one call's attendees – is covered in how to automate LinkedIn messages.

Guardrails

  • You approve every send. The prompts above say so, and the workflow never needs an exception.
  • Pace is enforced, not suggested. Per-action limits and working hours apply whatever Claude is asked to do.
  • One account, one rep. Each rep connects their own account; sharing one account across a team mixes threads and sender identity.
  • Claude reads what your account can see. Private profiles stay private, and nothing here returns emails or phone numbers.
  • The CRM stays the system of record. The brief and the follow-up feed it; they do not replace your CRM or your call recorder.

Frequently Asked Questions (FAQ)

Yes, in two ways. LinkedIn's own connector in Claude's directory, LinkedIn Hiring Pro, drafts and posts one free job and does nothing else. To work in your own account – read profiles, see who you know at a company, read and send messages – connect a tool on your account, such as Linked API as a custom connector, its MCP server, its CLI or its skill. See the LinkedIn MCP server overview.

Public web research and connected data providers can supply some LinkedIn information. The plugin's account research uses web search and connected data providers, and its call prep uses CRM titles. None of the connectors in Anthropic's sales materials links your own LinkedIn account, so your connections, your message threads and replies on LinkedIn are outside it.

Yes, when the connected account has a Sales Navigator subscription and a Linked API Plus seat. Claude can then read Sales Navigator profiles and companies and send Sales Navigator messages through the Sales Navigator tools listed in available tools.

Not with this workflow: the prompt requires your explicit approval before anything is sent. The per-action limits and working hours on your account apply either way.

Yes. The same steps run in Claude Code through the MCP server, the CLI or the LinkedIn skill. For a marketing workflow in Claude Code, see Claude Code for LinkedIn marketing; for finding and qualifying new prospects before there is a call, see AI for sales prospecting.

Build it in, or hand it to an agent

Build the reply routing and the brief into your own tools with the REST API, the Node and Python SDKs or the CLI. Or have Claude run it out of the box through the MCP server, the agent-friendly CLI or a ready-made skill. To map everyone involved in a deal, see mapping a B2B buying committee; to watch for activity between calls, LinkedIn buying signals. Plans start at $49 a seat a month billed annually – see pricing.

Facts verified October 29, 2026 – Anthropic's sales page, Sales plugin and Claude connector directory checked on that date.