Connect an AI assistant with MCP

Open the MCP page and add the server to your assistant

Everything starts at the MCP page in our developer docs. It carries the server URL, a short setup path per client, and the full reference for what the connection can do. The server address is printed right in the header of that page, and it is the only thing you ever hand your assistant. There is no API key to copy and nothing to paste into a config file.

The setup is a few clicks wherever you are working:

• Claude: Mailsoftly is listed in the Claude directory, so the whole setup is one click. Open the Mailsoftly listing in the directory (or the Claude integration page) and press Connect. Claude opens Mailsoftly in your browser; sign in, or create a free account right there, read the permissions the app is asking for, and approve. You land back in Claude with the connector ready. Adding it by hand as a custom connector with the server URL also works, but the directory route is the one to use
• ChatGPT: open the settings area where connectors and apps are managed, choose to add one by URL, and paste the same server URL. ChatGPT walks the same sign-in and consent flow
• Anything else: any MCP client that supports remote servers with OAuth works the same way. Give it the server URL and let it discover the rest

There is also a local server you run yourself, meant for developers who would rather host the process on their own machine. It needs Node.js 20 or newer and a Mailsoftly API key, it plugs into Claude Desktop, Cursor or VS Code through their MCP config files, and it exposes one tool per live API endpoint instead of the curated set. The same page covers it end to end.

One URL is the whole setup. The next screen you see is ours.

Open the MCP page

Check what the assistant will actually be able to do

This is the part worth reading before you approve anything. The hosted server publishes 134 curated tools rather than every API operation, so the assistant reaches for the right one instead of guessing between near-identical endpoints. Most of them only read: contacts, lists and who is on them, tags, custom fields, segments, the suppression list, forms and their submissions, saved templates, automations, landing pages, SMS campaigns, webhook endpoints, signatures, team members, sender addresses, campaigns with their results (per-link clicks, per-recipient engagement, the audience a draft would reach), each contact’s activity, imports in progress, files and images, content blocks, the brand kit, domain health checks, connected apps and the audit log, and the account itself with its plan, limits and sending state.

The tools that change something fall into five groups, and the sending and automation groups are the ones that reach real people:

• Contacts, lists, tags, templates, forms, landing pages and webhook endpoints: create, edit and remove them. Deletes are flagged so the assistant asks you first, and campaigns and pages go to a reversible trash instead
• Data and assets: import_contacts loads a CSV into a list, add_contact_note and update_contact_subscriptions work on one person, upload_file and upload_image add assets your campaigns link to, and invite_team_member or update_team_member_role manage the team behind their own permission
• Drafts: draft_campaign writes a campaign into your account as a draft from a text brief, a saved template or HTML you bring; update_campaign edits it in place; create_automation builds an automation as a draft. None of these deliver anything
• Sending: send_campaign delivers a draft now and schedule_campaign at a later moment. Real mail reaches real people and there is no unsend
• Automations: update_automation_status switches an automation on or off, and enroll_contact_in_automation starts an active automation for one contact. Activation needs a verified account

send_campaign, send_test_email, unsubscribe_contact and every delete are flagged to your assistant as destructive, which is the signal well-behaved clients use to ask you before running them. Every sending and activating tool sits behind the separate sending permission, behind the account’s trust tier, and behind the plan: a free, unverified account may keep a small number of automations, forms and pages, and the assistant is told so with an upgrade or verify path instead of an error. There is still no tool that touches billing, creates or revokes an API key, exports your contacts or changes account settings; those simply are not published.

Read this table once and you know the outer edge of what any connected assistant can reach.

Approve the permissions on the consent screen

After you sign in, Mailsoftly shows you its own consent screen. The heading names the app that is asking, the subtitle names the company it wants to work with, and under "This app will be able to:" you get one plain-language line per permission it requested. Nothing is granted by default, and an app only gets what it asked for and you approved.

The eight permissions read like this:

• Read your contacts, lists and tags
• Create and update contacts, lists and tags
• Read your campaigns and their reports
• Create and edit campaign drafts
• Send or schedule campaigns to your contacts
• Add addresses to your suppression list
• Read your account and company profile
• Invite teammates and change their roles

Sending is deliberately not in that list. When an app asks for it, it appears instead as a separate checkbox that starts unchecked, labelled "Allow this app to send or schedule campaigns", with the hint that everything else keeps working without it. Leave it off and the assistant can still research, draft and prepare; you press send yourself. Tick it only when you actively want the assistant to put mail in front of your contacts. The box starts unchecked every single time, including on a reconnection, so an earlier yes never carries over silently.

Press Allow access to finish, or Cancel to walk away. A footnote at the bottom shows the address you will be sent back to, which is worth a glance: it should belong to the app you are connecting.

Approve, and you land back in your assistant with the connection live.

Turn the connection off whenever you want

A connection is not a one-way door. Open Settings, find the Advanced group, and choose Connected apps. Every third-party app holding live access to your company is listed there with Authorized by, Last authorized, and the exact permissions it was granted, written in the same words you read on the consent screen. Review connected apps and access keys reads that screen column by column, alongside the API keys your own systems hold.

To cut one off, use the Revoke access button on its row and confirm. Access stops right away, for everyone in the company rather than only the person who connected it. A few things worth knowing:

• Connected apps is admin only, to see and to revoke, which matches the gate on the consent screen
• Access tokens are short lived and refresh tokens rotate, so a token that leaks somewhere stops being useful quickly, but revoking here is what makes it immediate
• The connected app is recorded on the contact activity it creates, so you can look back later and see which app did what

If you would rather build against the API yourself instead of using an assistant, start with Use the Mailsoftly API, and Connect integrations and the API covers the ready-made connections and Zapier.