What it is
The whole loop an SDR runs in a morning, written down so an AI client can run it with you: work out the search, find the people, research the ones worth the time, and draft an opener that quotes something real. It ends at a file of drafts. Nothing in this kit can send a message, and the endpoint it talks to exposes no tool that could.
Who it is for
- A founder or an SDR doing their own prospecting, who already uses Claude, Cursor or VS Code and wants the research half automated without handing over the account.
Capabilities
Each of these is one atomic thing the product can do, and each names what performs it. A capability fulfilled by a tool maps onto a tool the MCP endpoint serves today — follow the link and you can read the exact request it makes and what comes back.
List the sending account's network
MCP toolMedium riskFirst-degree connections of the connected sender account, up to two hundred per call. It is that account's network, not the network of whoever is signed into the app — which is exactly the distinction that matters when the sender is a dedicated account.
Your LinkedIn session · list_linkedin_connections
Read what someone has posted
MCP toolMedium riskUp to twenty recent posts by one member, newest first, with reposts flagged. An empty array — not an error — is the normal answer for a senior buyer who does not post, and a workflow has to handle that as a case rather than as a failure.
Your LinkedIn session · get_linkedin_posts
Read one profile in depth
MCP toolMedium riskOne call returns identity, current role and the date it started, follower count, the premium flag and the latest post. Every enrichment field is best-effort: a profile that hides its activity still returns its identity, with the post fields simply absent.
Your LinkedIn session · get_linkedin_profile
Find people matching an ICP
MCP toolMedium riskRuns an ICP description through a connected account's LinkedIn search and returns up to 50 profiles per call. Profiles LinkedIn masks as "LinkedIn Member" because they are out of network are dropped rather than returned with a blank name. Typed Sales Navigator filters are used only when the account has a Navigator seat; otherwise every criterion is folded into one keyword string.
Your LinkedIn session · search_linkedin_people
Read the approval queue
MCP toolLow riskThe AI-written replies waiting for a person to press send. An agent can read and summarise this queue; it cannot approve anything in it, and that is deliberate — sending is the one irreversible action in the product.
Read only · list_tasks
Read the unified inbox
MCP toolLow riskOne thread per lead across every sender account, with full message bodies and — once a reply has been classified — its intent. There is no since or limit argument; the whole inbox comes back.
Read only · get_inbox
Read lead lists
MCP toolLow riskThe lead lists and the leads in them, with whatever enrichment was stored at import. The whole set comes back in one payload: there is no paging or filter argument yet, so a large workspace returns a large response.
Read only · list_leads
See which LinkedIn identity is available
MCP toolLow riskThe accounts this workspace can act as, each with status, warm-up stage, how much of today's invite and message budget is spent, and a health score. Reads our own database, so it costs nothing against the LinkedIn rate limit and is the cheapest first call an agent can make.
Read only · list_sender_accounts
Check the remaining LinkedIn budget
MCP toolLow riskHow many LinkedIn tool calls this workspace has made this month, broken down by tool, plus the rate limit in force. The call an agent makes to pace itself instead of discovering the limit by hitting it.
Read only · get_usage
Draft a follow-up
AI skillLow riskWrites the next message in a thread from what the person actually replied, including the version that says nothing further is coming. It drafts; a person decides whether it goes.
Human approves
Turn an offer into search filters
AI skillLow riskConverts a written offer and a few existing customers into the arguments search_linkedin_people actually takes — keywords, location, industry, company, exclude_company — and says which of them LinkedIn will treat as a typed filter versus fold into keywords.
Read only
Draft a first message
AI skillLow riskWrites an opener grounded in a cited line from the person's profile or a post, in the project's voice, under LinkedIn's 300-character invitation-note limit. It produces text in a file. Nothing in the kit can send it.
Human approves
Write a brief on one person
AI skillLow riskCompresses a profile and its posts into a one-screen brief in which every claim names the call that produced it. Anything the tools did not return is written as "not visible" rather than inferred, because an invented detail in a first message is worse than a missing one.
Read only
Sort replies by intent
AI skillLow riskClassifies an inbox into the same seven intents the app itself uses — meeting, interested, question, objection, not_now, not_interested, neutral — so the queue is read in the order that matters instead of newest-first.
Read only
Find a warm path in
AI skillLow riskIntersects the sending account's first-degree connections with a target's company to tell you whether there is an introduction to ask for. It answers from the sender's network only; it cannot see anybody else's connections.
Read only
One person, fully researched
WorkflowMedium riskResolve the profile, read it, read up to ten of their posts, check for a warm path, and produce a dossier with the calls that produced each claim listed underneath it. Four LinkedIn calls per person, and the workflow says so before it starts.
Your LinkedIn session
A day of prospecting, ending at a stop
WorkflowMedium riskBudget check, one search, dedupe against the sender's existing connections, research the shortlist, draft an opener each, write them to a file, and stop. The workflow has no send step and the endpoint exposes no tool that could add one.
Human approves
Write a researched lead into a CRM
Planned — not available yetLow riskNot built yet. Research comes out as files you own. There is no CRM write path, so the last step is still a copy-paste or an import.
Official API
One capability is marked planned: nothing fulfils it today. It is listed because the roadmap should be visible, and it does not run when you install this.
A worked example
Lifted from the kit’s own EXAMPLES.md — list_sender_accounts — free, always first. Field names are real. Values are illustrative — they are the shape you will get, not data from anyone's account.
{ "jsonrpc": "2.0", "id": 1, "method": "tools/call",
"params": { "name": "list_sender_accounts", "arguments": {} } }{
"accounts": [
{
"id": "snd_02",
"name": "Marcus Feld",
"status": "active",
"connected": true,
"warmupStage": 4,
"dailyInviteLimit": 24,
"dailyInviteSent": 11,
"dailyMsgLimit": 40,
"dailyMsgSent": 6,
"health": { "score": 92, "level": "good" }
}
],
"default": "snd_02"
}What is included
- An agent definition that keeps a tool budget and refuses to invent details
- A daily workflow with the exact call sequence and its cost in LinkedIn calls
- Opener and follow-up rubrics with the 300-character invitation limit built in
- A reply-triage pass over the unified inbox, sorted by intent rather than by date
- Prompts, worked examples with real response shapes, and the endpoint's verbatim error list
The files
| File | What it holds |
|---|---|
| README.md | What it is, what it needs, what it cannot do, and what is in the directory. |
| INSTALL.md | The exact commands for Claude Code, Claude Desktop, Cursor and VS Code, plus a call that proves the connection before you trust it. |
| AGENT.md | The agent definition to paste into your client's project instructions: role, tool budget, rules, and where it stops. |
| SKILL.md | The judgement written down — the rubric, what counts as evidence, and what to write when the evidence is not there. |
| WORKFLOW.md | The run step by step, with the exact tool calls, the cost of each, and the human gate it ends at. |
| PROMPTS.md | Prompts to paste, each one naming the tools it will spend. |
| EXAMPLES.md | Real request and response shapes from the endpoint, with illustrative values and the field names that are not. |
| TROUBLESHOOTING.md | The errors this endpoint actually returns, quoted verbatim, and what each one means. |
| .env.example | The values the install asks for, in one place. Nothing in the kit reads it by itself. |
What it cannot do
Printed before you pay, not after you ask for a refund.
- It does not send. No invitation, message, comment or connection request leaves this kit, because no tool on the endpoint sends anything.
- A run of twenty-five prospects spends about fifty LinkedIn calls, which is two rate-limit windows: roughly ten minutes of wall clock, paced.
- Search quality is the account's search quality. Without a Sales Navigator seat, location and industry become keywords rather than typed filters.
- It reads no CRM and writes to none. Output is Markdown and CSV in your own directory.
What a run costs: 1 search + 2 calls per prospect researched. Twenty-five prospects ≈ 51 LinkedIn calls ≈ two rate-limit windows. The endpoint allows a fixed number of LinkedIn calls per five minutes per workspace, which is documented per tool.
Install
These are the commands from the kit’s INSTALL.md, read off disk at build time — so this page and the file you download cannot disagree. Pick the client you work in.
Claude Code
claude mcp add --transport http topclozer https://topclozer.com/api/mcp \
--header "Authorization: Bearer tcz_live_YOUR_KEY"Claude Desktop — custom connector
Settings → Connectors → Add custom connector. That dialog has no header field, so the key goes in the URL:
https://topclozer.com/api/mcp?key=tcz_live_YOUR_KEYClaude Desktop — older builds, via mcp-remote
claude_desktop_config.json:
{
"mcpServers": {
"topclozer": {
"command": "npx",
"args": [
"-y", "mcp-remote",
"https://topclozer.com/api/mcp",
"--header", "Authorization: Bearer tcz_live_YOUR_KEY"
]
}
}
}Cursor — .cursor/mcp.json
{
"mcpServers": {
"topclozer": {
"url": "https://topclozer.com/api/mcp",
"headers": { "Authorization": "Bearer tcz_live_YOUR_KEY" }
}
}
}VS Code — .vscode/mcp.json
{
"servers": {
"topclozer": {
"type": "http",
"url": "https://topclozer.com/api/mcp",
"headers": { "Authorization": "Bearer tcz_live_YOUR_KEY" }
}
}
}Prove it before you trust it
curl -s https://topclozer.com/api/mcp \
-H "Authorization: Bearer tcz_live_YOUR_KEY" \
-H "Content-Type: application/json" \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'Compatibility
| Target | Supported |
|---|---|
| Claude | Yes |
| Cursor | Yes |
| VS Code | Yes |
| Run here | No — hosted seats are a separate tier |
| Download | Yes |
What you need before it runs
- A TopClozer workspace API key
- A LinkedIn account connected to that workspace — a dedicated one, not your personal profile
- An MCP-capable client: Claude Code, Claude Desktop, Cursor or VS Code
LinkedIn plan: A free LinkedIn account is enough. A Sales Navigator seat on the connected account turns location, industry and company into typed filters instead of keywords, which materially changes who the search returns.
Security and risk
Medium risk. Acts through your own session or writes with a human approving each action. Restriction is unlikely but not impossible.
Human approves. Drafts the action; nothing irreversible happens until a person presses send.
The live MCP endpoint is read-only by design, and the outreach engine keeps every irreversible action behind a human approval gate. None of that is LinkedIn endorsing any of this, and nobody can promise your account will not be restricted — least of all a vendor selling you automation. What we store and who can read it is a separate page, and it is the one that answers that question.
Provider: TopClozer — a verified seller: we wrote it, we maintain it, and the support request comes to us.
Pricing
$49 — A skill plus MCP wiring, workflows and the files they read. The whole ladder is on the pricing page.
No payment processor is connected to this site, so nothing here can charge a card — and a button that looked like it could would be the lie this catalogue exists to avoid. What happens instead: the order button writes a real pending purchase against your workspace and hands you its reference, a person invoices you, and marking that order paid is what puts the files in your library. Slower than a card, and every step of it is real.
Version history
v1.0.02026-09-20
First release. Daily prospecting workflow, brief and opener rubrics, reply triage, and the endpoint's verbatim error list.
Reviews
Nothing here. A review appears on this page only when it is attached to a verified purchase of this product, and none has been published. An empty section is the correct state for a catalogue this young — the alternative is the kind of testimonial that means nothing to the person reading it.
Related products
Company Research Kit
Map an account from its people, because nothing here reads a company page.
Capabilities
$49
LinkedIn Starter Agent
Connect the endpoint and get one real answer in five minutes. Free.
Capabilities
Free
Prospect Research Kit
One screen on one person, with every claim traced to the call that returned it.
Capabilities
$49