· 9 min read · By Marwene Amor
lemlist ships a first-party MCP server. Connect it to Claude and you can ask, in plain language, for a lead list, an enrichment pass, a sequence, or the numbers on a campaign that is underperforming, without opening lemlist at all.
That is the useful half. The other half is the part the product pages leave out, and it decides whether this belongs in your workflow. Both are below, along with the three different ways to connect and why they do not expose the same thing.
MCP, the Model Context Protocol, is an open standard for letting an assistant call external software. lemlist runs a server at https://app.lemlist.com/mcp that exposes its own features as callable actions, covering campaign operations, lead management, database sourcing, email discovery and verification, analytics, and webhook events.
How many tools you actually see depends entirely on how you connect, and the numbers in circulation disagree because they describe different things:
| Route | Tools exposed | Per-tool permissions | Best for |
|---|---|---|---|
| Claude connector directory | 10 consolidated | No | Chat use, one-click install |
| Manual wizard from the lemlist dashboard | 34 raw | Yes, Allow / Ask / Block | Operators who want granular control |
| Claude Code or Desktop config | Server endpoint direct | Client-dependent | Developers, scripted setups |
lemlist's marketing pages describe more than forty actions. Ten coarse tools rather than thirty-four raw ones is the right call for conversational use, so the smaller number in the directory build is a design decision rather than a downgrade.
lemlist publishes a Community connector in Claude's public connector directory. Installing it takes one click and skips the manual URL step entirely. As of our check in July 2026 it exposes ten consolidated tools: campaign creation with a full sequence, campaign details and sequences, inbox conversations, lead search with filters, pushing leads to contacts, settings, a skill loader, and help-center search.
This is the route for anyone who wants to use lemlist from a chat window and stop thinking about it.
The lemlist dashboard carries a "Run lemlist from your AI assistant" panel with connect cards for Claude and ChatGPT. The wizard hands you the MCP URL, opens Claude's Add custom connector form with the name and URL pre-filled, walks the OAuth handshake with a team selector so you scope the grant, and ends on a permissions panel.
That panel is the reason to take this route. Each of the thirty-four tools is set to one of three levels:
Those defaults are well chosen. Nothing that sends or deletes happens without you seeing it first.
In Claude Code, one command:
In Claude Desktop, an entry in claude_desktop_config.json:
Both use OAuth, so nothing sensitive sits in a config file. Access tokens last an hour, refresh tokens thirty days. If your client needs a header instead, which is the case for Cursor, generate a key in lemlist under Settings, Team, Integrations and send it as X-API-Key.
Claude shows an "Unverified third-party app" warning during the OAuth flow. This is standard for every third-party MCP and says nothing about lemlist specifically, but it does stop non-technical users cold the first time they see it.
The grant is coarse. OAuth asks for full API access equivalent to an API key, scoped to the team you pick. The per-tool Allow, Ask and Block screen compensates partly, but the underlying authorisation is account-level. If you administer lemlist for clients, scope the team before you approve rather than after.
Research and assembly, not sending.
Asking for a filtered list and getting it back without building the query in a UI saves real time when you are exploring a segment rather than executing a known one. Enrichment in the same breath as the search removes an export-import round trip. Reading campaign performance and asking which step is leaking is faster in conversation than in a dashboard, because the follow-up question costs nothing.
There is also a competitive angle worth stating plainly: as of July 2026, neither Instantly nor Smartlead ships a first-party MCP. If your daily work is drifting toward assistants, that gap matters more than most headline feature comparisons. We cover the wider picture in lemlist vs Smartlead vs Instantly.
Nothing in this setup reacts to an event. A reply arriving at eleven at night, a bounce, a new inbound lead: none of them wake anything up. The connection acts when you prompt it, and only then.
That single property rules out most of what people picture when they hear autonomous outreach. No automatic reply drafting when a prospect answers. No automatic enrolment of a new lead into a campaign. No overnight operation. Booking the meeting still ends with you clicking something.
This is not a flaw in lemlist's implementation. It is what the protocol is for: an assistant calling tools during a conversation. Event-driven behaviour needs a scheduler or an automation layer sitting on top, and that is a different project with a different cost.
One more constraint in the same family: lead-database operations consume lemlist credits at the same rate as doing them by hand. Email finding, verification and phone enrichment all draw the same balance. A conversational interface makes it very easy to burn a month of credits in an afternoon without noticing, because asking costs you nothing and the request looks the same whether it returns ten rows or a thousand.
lemlist's developer documentation describes the sourcing database as 450 million-plus prospects. Its product page for the same feature says 600 million-plus. Both are lemlist's own pages. Treat database-size claims from any vendor as marketing rather than a specification, and judge the tool on what a search actually returns for your segment.
Yes, if you already pay for lemlist, you already work inside Claude, and your bottleneck is research and list assembly. Minutes of setup remove a lot of tab-switching.
No, if you are hoping to replace a person or run campaigns unattended. You will feel the gap in week one.
An assistant that can launch a campaign in ninety seconds is worth nothing if the domain sending it fails authentication. Speed at the top of the funnel does not survive a spam folder at the bottom.
Read your own live records with our free deliverability checker, which pulls SPF, DKIM and DMARC straight from public DNS. We audited 156 B2B lead generation agencies and 56 percent had a real defect on their own domain: the full study has the method and the counts.
It is an endpoint lemlist publishes at app.lemlist.com/mcp that exposes lemlist's features as tools an AI assistant can call. MCP, the Model Context Protocol, is an open standard for connecting assistants to external software. How many tools you see depends on how you connect: the Claude directory connector exposes ten consolidated tools, the manual wizard exposes thirty-four raw ones, and lemlist's marketing pages describe more than forty actions.
Three routes. The simplest is lemlist's Community connector in Claude's public connector directory, which installs in one click. The manual wizard from the lemlist dashboard gives you per-tool Allow, Ask and Block permissions instead. For Claude Code, one command does it: claude mcp add --transport http lemlist https://app.lemlist.com/mcp.
Because the connector is published by lemlist rather than by Anthropic. This is standard for every third-party MCP and is not specific to lemlist. It looks alarming the first time and it is not a signal about lemlist's security.
The OAuth grant is coarse: full API access equivalent to an API key, scoped to the team you select. What compensates is the per-tool permission screen in the manual wizard, where each tool is set to Allow, Ask or Block. The defaults are sensible: read operations such as Get Campaigns run silently, action operations such as Send Email ask for confirmation, and destructive operations such as Delete Lead are blocked.
Yes, for anything that touches the lead database. Email finding, verification and phone enrichment draw on your lemlist credits exactly as they would if you ran them by hand. Asking an assistant instead of clicking does not make a lookup free, and a conversational interface makes it easy to burn a month of credits in an afternoon.
No, and this is the limitation worth understanding before you build anything on it. The connection is on-demand: it acts when you prompt it. Nothing in it reacts to an event. A reply arriving at eleven at night, a bounce, a new inbound lead, none of these wake anything up. Event-driven behaviour needs a scheduler or an automation layer on top, which is a separate project.
Not as first-party integrations, as of our last check in July 2026. lemlist is the only one of the three that ships its own MCP server and publishes a connector. If your workflow is moving toward AI assistants, that gap matters more than most headline feature comparisons.
If you already pay for lemlist and you already work inside Claude, the setup takes minutes and removes a lot of tab-switching for research and list building. If you are hoping it will replace a person or run campaigns unattended, it will not, and the gap between the marketing and what the protocol allows today is wide.