You can now run most of your cold outbound from a Claude chat window. Smartlead has an MCP server that connects Claude directly to your campaigns, sending accounts and lead data, and this week the company is pushing the workflow hard: a step-by-step build guide published October 28 and a live "outbound A to Z, run from Claude" session for its whole user base on October 29. Here is what the connection actually does, how to set it up in about ten minutes, and where you still need to stay in the loop.
We run a growing share of our own operations from Claude (here is our hands-on walkthrough of Claude Cowork in the cloud), so this setup is close to home. The short version: it works, it saves real time on reporting and deliverability triage, and it is not an autopilot.
What does the Smartlead MCP server actually do?
MCP (Model Context Protocol) is the open standard that lets an AI assistant call real tools instead of just talking about them. Smartlead's MCP server registers its toolkit inside Claude, over 116 tools as of October 2026 according to Smartlead's own build guide, covering three big areas:
- Campaign performance: live stats, reply rates, sends, bounces, side-by-side summaries.
- Deliverability: spam diagnostics, warmup status, sending-account health and risk flags.
- Lead and account data: who replied, who is still in sequence, sender inventory.
Instead of tab-hopping between dashboards, you ask: "Which campaigns lost reply rate this week, and is it the copy, the list or the accounts?" Claude queries Smartlead and answers with your real numbers.
Two caveats, both from Smartlead's help center, as of October 2026: the server runs over SSE transport only, and it works in the Claude Desktop app (Mac and Windows), with Claude web support planned but not live yet.
How do you connect Smartlead to Claude?
The whole setup is one API key and one config file. No code.
- Generate your Smartlead API key. In the Smartlead dashboard: Settings, then API Keys, then Generate API Key. Treat it like a password.
- Install Claude Desktop if you have not already (any Claude account works with the desktop app; Smartlead does not name a required plan).
- Add the server to Claude's config. In Claude Desktop: Settings, then Developer, then Edit Config. In claude_desktop_config.json, add a "smartlead" entry under mcpServers that runs npx -y mcp-remote against https://mcp.smartlead.ai/sse?user_api_key=YOUR_API_KEY. You need Node.js installed for npx. If you already have other MCP servers, add the block inside the existing mcpServers object.
- Restart Claude Desktop and toggle Smartlead on under the search-and-tools icon in a new chat.
- Test it. Ask Claude to check the health of every sending account and flag problems. If the answer contains your real account data, you are connected. A generic answer means you are not; recheck the config file and restart.
What should you run first?
These five prompts are where the connection earns its keep in week one:
- Morning account health audit. Status, warmup, daily limits, bounce rate and risk flags across every sending account, in one table.
- Reply gap check. Find leads who replied but are still sitting in active sequences, with campaign and step. This one finds money.
- 30-day campaign summary. Sends, opens, replies, bounces and unsubscribes, sorted by reply rate.
- Pre-launch check. Sequence setup, account health and daily limits verified before you hit go.
- Deliverability triage. Reply rate dropped with no copy change? Have Claude separate account problems from campaign and list problems before you touch anything.
Where do humans stay in the loop?
Smartlead's own guidance keeps a person at four checkpoints, and we agree: review what the audit flags, approve any lead removals or handoffs, fix what the pre-launch check catches, and launch campaigns yourself. Claude gets you a strong first pass fast. It does not replace judgment on list quality, offer and copy, and you should not let it send without review until you have watched it work for a while.
Should you set this up this week?
If you already pay for Smartlead: yes. The MCP server uses the API access your plan already includes (usage counts against your plan's API quota), so the only cost is ten minutes of setup. Smartlead has been shipping quickly this quarter; it added auto-dialing earlier this month, and the MCP push is part of the same run.
If you are still choosing a sending platform, MCP support belongs on your checklist now, right next to the real cost of your cold email stack. An outbound tool your AI assistant can operate is worth more than one it cannot.
And if you want the wiring done for you, custom connectors included, that is a thing we build: see MCP connectors.
Frequently asked questions
Do I need to know how to code to connect Smartlead to Claude?
No. The setup is generating an API key and pasting one block into Claude Desktop's config file. The only prerequisite beyond that is Node.js, which the connector uses to run.
Does the Smartlead MCP server work in the Claude web app?
Not yet. As of October 2026, Smartlead's help center says the server works with Claude Desktop on Mac and Windows, with web support planned.
Does connecting Smartlead to Claude cost extra?
There is no separate fee for the MCP server. Calls count against the API quota included in your Smartlead plan, and heavy use can hit endpoint limits on lower tiers.
Can Claude launch campaigns on its own?
No, and it should not. The workflow Smartlead recommends keeps a human on launches, lead removals and fixes. Claude drafts, audits and diagnoses; you decide.
What data can Claude actually pull from Smartlead?
Campaign stats, deliverability and spam diagnostics, and lead, account and sender data. Example tools include get_campaign_status, fetch_lead_data and check_deliverability.
If you want this whole system designed, built and run by operators (lead sourcing, copy, deliverability and the agent wiring included), that is exactly what The Prospector, our prospecting agent build, is for.
