Lemlist MCP Review 2026: Features, Setup and Limitations

Key takeaways:

• Lemlist’s MCP server sits at https://app.lemlist.com/mcp and defaults to OAuth, so there is no API key to create, store or rotate.
• Tokens are handled for you, with an access token lasting about an hour and a refresh token about 30 days. Leave it unused for a month and you will be asked to reconnect.
• The limits that bite are set by your AI client rather than by Lemlist: Claude’s free plan allows only one custom connector, and full MCP is unavailable on ChatGPT web for Plus and Pro users.
• MCP is listed as a feature of the entry Email plan, which starts at $55 per month billed yearly.

Lemlist has written the most thorough MCP setup documentation we have read, and the most useful part of it is about other companies’ products.

The documentation covers which AI client plans block the connector, which ones need an admin to approve it, and what to do when a refresh token quietly expires. That is the kind of detail most vendors leave their customers to discover in a support ticket.

This review covers the server, the setup, the client-side gates and the limits. Multichannel outreach is a category we compete in, and we run a connector of our own, so weigh the comparison accordingly and take the factual sections from the documentation they come from.

What Lemlist MCP Is

Lemlist MCP is a hosted Model Context Protocol endpoint, built on the specification assistants use to operate third-party software directly. Lemlist’s framing is that you can manage campaigns, search leads and analyze performance through conversation.

The endpoint is https://app.lemlist.com/mcp, and the documented use cases run to auditing campaigns, creating or updating sequences and managing leads without switching tools.

The platform underneath is multichannel rather than email-only, which matters for what the server can eventually reach: Lemlist covers email, LinkedIn automation, SMS, a WhatsApp add-on and a built-in call dialer, alongside a leads database it describes as 650M+ records. Our Lemlist review covers that platform in detail, and the server is best understood as a window onto it rather than a product of its own.

Two Ways In, and One Is Better

Lemlist supports OAuth and API keys, and it recommends OAuth. That recommendation is worth following.

With OAuth there is no key to create or store. Your client completes authentication through the browser, including consent, token exchange and refresh. On first use a consent page opens where you select your team and authorize access, which is a nice touch for anyone who belongs to several.

The alternative is an API key generated under Settings, then Team, then Integrations. It exists for environments that cannot complete an OAuth flow, and it puts a long-lived secret back into your configuration.

The mechanism behind the recommended path is worth naming because it is what makes browser-based auth safe on a desktop client. Lemlist’s tooling completes the exchange using PKCE, defined in IETF RFC 7636, which exists because OAuth 2.0 public clients using the authorization code grant are vulnerable to having that code intercepted. PKCE binds the code to the client that requested it, so intercepting it is not enough to steal the session.

Token Lifetimes, and the Failure They Cause

This is the section most likely to save you a support ticket later.

Lemlist manages tokens automatically, with an access token lasting roughly one hour and a refresh token roughly 30 days. In normal use you never think about either.

The failure mode appears when you stop using it. Lemlist documents that if tools suddenly fail or ask you to reconnect, the likely causes are a refresh token that expired after about 30 days of inactivity, access that was revoked, or your removal from the team you originally authorized.

That behavior follows from how the underlying framework works. RFC 6749 defines the OAuth 2.0 authorization framework and its refresh-token mechanism, where a longer-lived credential obtains fresh access tokens without the user re-authenticating. Expiry on that longer-lived credential is a deliberate security property rather than a bug, and reconnecting is the intended remedy.

The practical consequence is worth planning for: a Lemlist connection used monthly rather than weekly will need periodic reconnection, which makes it a poor fit for a fully unattended agent unless something keeps the session warm.

How to Set Up Lemlist MCP

Lemlist’s client-by-client instructions are the most detailed we have read, even where a rival covers more ecosystems. The OAuth route is the same shape everywhere: point the client at the endpoint and authorize in the browser.

Claude

In Claude Desktop, open the sidebar, click the + next to Connectors, choose Add custom connector, and set the name to lemlist and the URL to https://app.lemlist.com/mcp. The consent page opens on first tool use.

Claude Code takes the same endpoint from the command line, and mcp-remote handles the full flow: discovery, client registration, browser consent, token exchange and refresh.

Scope is the option most people miss. Claude Code defaults to --scope local, meaning the server exists only in the project where you ran the command. --scope user makes it available across all your projects, and --scope project shares it with your team through .mcp.json in the repo, where each teammate still authenticates individually.

Cursor, ChatGPT and Codex

Cursor connects over OAuth in the same way. ChatGPT is where it gets complicated, and the complication is ChatGPT’s rather than Lemlist’s.

On ChatGPT web, MCP lives behind Apps in developer mode, and full MCP is not available to Plus and Pro users at all. Lemlist’s documented workaround is to use the ChatGPT desktop app or the Codex route instead, where there is no plan gate.

The Client-Side Gates Nobody Mentions

The constraints below are the ones most likely to stop you, and none of them is under Lemlist’s control.

If tools are slow to appear in Codex, Lemlist suggests raising startup_timeout_sec to 20 and tool_timeout_sec to 60 in the server block. And in Claude Code, claude mcp list confirms whether the server was actually added, which matters because a default local scope means it exists in one project only.

Lemlist MCP Pricing

Lemlist lists lemlist MCP as a feature of the entry Email plan, which puts it on the generous end of the field. A couple of vendors do gate MCP behind a paid add-on, Outreach and Salesloft among them, and most others include it with any paid tier as Lemlist does.

Yearly billing saves 20% against monthly and quarterly saves 10%, so the $55 and $87 figures are the discounted ones. A 14-day free trial covers both self-serve plans.

The plan boundary is the thing to understand before buying. The Email plan carries the leads database, the email and phone finder, the unified inbox, deliverability and warm-up, CRM integrations and the API. LinkedIn automation, SMS, the WhatsApp add-on and the call dialer all sit on Multichannel, so a team whose sequences depend on a LinkedIn touch is on the higher plan regardless of sending volume.

Enrichment is metered separately in credits, at one cent per credit, with an email costing 5 credits and a phone number 20. That works out around $10 for 200 emails or 50 phone numbers.

That plan split is worth modeling honestly, because the channel you need decides the tier rather than the volume you send. A team on the Email plan paying $55 for unlimited users is getting an unusually good deal; the same team needing one LinkedIn step is on Multichannel at $87 per user.

Comparing that against dedicated LinkedIn outreach tools is worth doing before committing, since a specialist plus an email plan sometimes costs less than one multichannel seat.

Where Lemlist MCP Falls Short

Most of the friction here is documented, which is more than most vendors manage, but documented friction is still friction.

The client-side gates are the biggest practical limit and they will affect a real share of readers. Being unable to use the connector on ChatGPT web with a Plus subscription is not Lemlist’s doing, and it is still a wall you hit.

Token expiry makes unattended automation awkward. A 30-day refresh window is a sensible security posture and a poor fit for an agent that runs monthly.

The Claude Code default scope catches people out. Installing the server in one project and then wondering why it is missing in another is the documented confusion, and the fix is knowing about --scope before you need it.

Tool availability is not fully enumerated in the setup documentation. Lemlist’s own advice is to ask the assistant what tools are available, and to expect that list to change over time, which is honest and less useful than a published catalog when you are planning a build.

The channel gap in the server is the one to check against your own motion. Campaigns, leads and performance are documented; configuring a LinkedIn step or a call task through the assistant is not, even though the platform supports both. If you run genuinely multichannel outreach, expect the conversational surface to cover reporting and lead management while the channel configuration stays in the interface.

What Teams Actually Use It For

Three jobs account for most of what Lemlist documents, and they show where the connector genuinely pays for itself.

Campaign auditing is the most immediately useful. Asking which campaigns are underperforming, or which sequences have not been touched in a month, is a question that takes real clicking in any interface and one sentence here.

Lead management is the second. Searching leads and updating them conversationally suits the tidying work that otherwise never gets done, because the friction of doing it properly exceeds the value of any single record.

Performance analysis is the third, and the one where a conversational interface genuinely beats a dashboard. A dashboard answers the question it was built for; an assistant answers the question you actually have, including the awkward follow-up.

Building a specialized agent is the documented power-user route, and Lemlist walks through creating one in Claude Code. That is where this stops being a convenience and starts being infrastructure, which is the same trajectory we see across AI-driven outbound generally: the connector arrives first, the agent that uses it properly arrives later.

What none of these do is replace the interface for building a sequence. Writing copy, setting conditional logic and reviewing how a multi-step sequence will actually unfold are still visual tasks, and no vendor we reviewed claims otherwise.

Lemlist MCP Versus Reply MCP

We compete with Lemlist directly and the two products are closer than most pairings in this cluster, so this section is the least neutral in the review. Judge it on the specifics.

Our facet here is multichannel branching, and it is where the two products genuinely differ in the tools they expose. Our server can attach a mailbox, a LinkedIn account and a sending schedule to a sequence, then start or pause it, which means the channel mix and the timing are both reachable from a conversation rather than only from the interface.

That matters because a real multichannel sequence is conditional. The LinkedIn step should fire when the email goes unopened, not on a fixed day, and expressing that branching is the difference between multichannel and merely multi-tool. We wrote up how that logic is built in our notes on conditional sequences, and it is the part of outbound least well served by any conversational interface today, ours included.

The authentication defaults are a mirror image worth noting. Lemlist prefers OAuth, which is frictionless and grants what your account can reach. We prefer a scoped API key, which is slightly more work and lets you grant an agent read access without write access. Neither is wrong; they optimize for different anxieties.

Where Lemlist is ahead is the entry price for multichannel breadth and the documentation itself. Unlimited users on the Email plan is a genuinely strong offer, and its client-gate documentation is better than ours. We charge separately for LinkedIn automation, at $69 a month for each account connected, so a small team comparing bundled pricing should count that in.

We should also be plain that our free route is a 14-day trial and nothing beyond it, since we run no permanently free plan.

Who Lemlist MCP Is For

Lemlist MCP suits existing Lemlist customers, and it suits developer-adjacent users best because the Claude Code and Codex routes are the best documented. If you already run Lemlist and use Claude daily, connecting it costs ten minutes and removes a tab.

Unattended automation is the poor fit, on token lifetime alone. ChatGPT-web-only teams are stuck too, though that gate belongs to OpenAI rather than to Lemlist.

Teams still choosing a platform should compare on channel coverage and price per seat rather than on the MCP server, since most vendors in this category now ship one. Our roundup of sales prospecting tools covers the field on those terms.

The MCP server is a reason to stay somewhere, not a reason to move. Every serious vendor now has one, they do broadly similar things, and the differences that will still matter in a year are channel coverage, deliverability and what a seat costs.

Comparing multichannel branching rather than connectors? Start a free trial and run the same sequence on both.