AI Assistants for Cold Email: What Connects and How
Which AI assistants connect to cold email infrastructure today: Claude, ChatGPT, Grok, Gemini, Cursor, Copilot, Le Chat and n8n, by task, with setup guides.
On this page
Short answer: in September 2026 every major AI assistant can reach outside tools, through MCP servers, API actions or both. What differs is where the server has to run and how it authenticates. With the Outreach2day REST API, the assistants that run on your computer (Claude Code, Claude Desktop, Cursor, GitHub Copilot in VS Code, Windsurf, Gemini CLI, Grok Build) work today through a small local server or plain API calls. So do ChatGPT (a custom GPT with an Action), Microsoft Copilot Studio (a custom connector) and n8n (HTTP Request tools). The web assistants that only accept a hosted MCP server (claude.ai connectors, grok.com, the Gemini app, Mistral Le Chat) wait for the hosted Outreach2day MCP server, which is in early access.
Checked September 2026. Each assistant's guide below cites the vendor documentation it was checked against.
Assistants at a Glance
Assistant | Integration feature (first available) | Works with Outreach2day today | Guide |
|---|---|---|---|
Claude (claude.ai, Desktop) | Custom connectors, remote MCP (1 May 2025); local servers in Desktop | Desktop, local server | |
Claude Code |
| Yes: API or local server | |
ChatGPT | GPT Actions; custom MCP connectors (4 June 2025) | Yes: GPT Action | |
Grok | xAI API remote MCP (Nov 2025); Grok Build (beta, May 2026) | Grok Build, local server | |
Gemini | Gemini CLI MCP (25 June 2025); Gemini app custom apps | Gemini CLI, local server | |
Cursor | MCP (0.45, January 2025); OAuth (1.0, June 2025) | Yes: API or local server | |
GitHub Copilot (VS Code) | MCP generally available in VS Code 1.102 (July 2025) | Yes: local server | |
Windsurf (Devin Desktop) | Cascade MCP (Wave 3, 13 February 2025) | Yes: local server | |
Microsoft Copilot Studio | MCP GA (29 May 2025); custom connectors | Yes: custom connector | |
Mistral Le Chat | Custom MCP connectors (2 September 2025) | Exports only, until the hosted server | |
n8n | MCP Client Tool (1.88.0, April 2025); HTTP Request tools | Yes: HTTP Request tools |
"First available" is the vendor's own date for the feature where it publishes one. xAI does not publish a date for custom connectors on grok.com, and Google does not publish one for custom apps in the Gemini app.
Assistants by Task
Four jobs come up in every cold email setup. What an assistant can do with each today, over the Outreach2day API:
Task | What the API allows | What we recommend an assistant does |
|---|---|---|
Buy domains, create mailboxes |
| Check names and size the setup; a person buys and orders |
Check warm-up and deliverability |
| Report daily; start or pause only after you approve the list |
Launch a campaign | Campaign endpoints exist but are not enabled for every workspace; sequencer export adds mailboxes to Instantly or Smartlead | Draft and lint the sequence; launch in your sending tool |
Triage replies and bounces |
| Group and draft; a person approves each reply (reply triage guide) |
By assistant, the practical split:
Assistant | Domains and mailboxes | Warm-up | Campaign copy | Replies |
|---|---|---|---|---|
Claude Code, Cursor, Copilot, Windsurf, Gemini CLI, Grok Build | Check names, plan from lead files | Live report | Draft and lint in files | Live digest |
Claude Desktop | Check names | Live report | Draft in chat | Live digest |
ChatGPT (GPT Action) | Check names | Live report | Draft in chat | Live digest |
Copilot Studio agent | Check names | Live report | Draft in chat | Live digest |
n8n | On request | Scheduled report | Not its job | Scheduled digest |
claude.ai web, grok.com, Gemini app, Le Chat | From exports | From exports | Draft in chat | Paste replies |
Pick and connect an AI assistant, step by step
Start from where you work. If your leads and copy live in a repo, pick the coding agent you already use. If they live in chat, pick ChatGPT or Claude Desktop. If the job is a daily report, pick n8n.
Create one API key per assistant. In the Outreach2day app, the API keys page. A key acts as your user in every workspace you belong to; one key per tool lets you revoke one without breaking the others.
Give read access first. Workspaces, domains, mailboxes, warm-up state and replies. Add writes (warm-up operations, replies) only with an approval step.
Put the sending rules in the assistant's instructions (list below), so its drafts and plans follow them.
Follow the assistant's guide in the table above, and compare its first answers with the app.
The Rules Every Assistant Should Follow
These are the numbers we use with customers; the cold email guide explains them:
At most 25 cold emails per mailbox per day, and 5 mailboxes per domain (about 125 cold emails per domain per day).
At least 14 days of warm-up before a mailbox sends cold email, with warm-up kept on while it sends.
Follow-ups after 2, 3, 5 and 5 days.
No links in cold emails, the signature included.
Heavy spintax on subjects when sending at volume, so the same subject line does not repeat thousands of times.
The same rules are packaged as agent skills in our skills library, for example the follow-up sequence planner. Each assistant's prompts page turns them into copy-paste prompts: Claude, Claude Code, ChatGPT, Grok, Gemini, Cursor and Le Chat.
What Changes With the Hosted MCP Server
The hosted Outreach2day MCP server is in early access. When it is generally available, the web assistants in the last row of the table (claude.ai, grok.com, the Gemini app, Le Chat) get live access too, and the local servers in the other guides become optional. Each guide on this page switches to the hosted setup at that point. Get early access on the home page.
Assistants We Left Out
Perplexity. Its help center documents local MCP servers in the Mac app, with remote servers "coming soon". We could not confirm a launch date or a custom remote connector, so there is no guide yet.
Zapier. Zapier is retiring Zapier Agents in favour of AI steps inside Zaps. We could not confirm from Zapier's documentation that those steps call a custom API as a tool.
FAQ
Which AI assistant is best for cold email?
For live data today: a coding agent (Claude Code, Cursor, Copilot, Windsurf, Gemini CLI or Grok Build) or Claude Desktop with a local server. For scheduled reports: n8n. For a team without developers: ChatGPT with a GPT Action or a Copilot Studio agent.
Can an AI assistant run my whole cold email setup?
It can plan, check, report and draft. Buying domains through the API is paid right away — give an agent a billing-scoped key only if you want it to buy. Sending replies emails real people, so a person should approve those.
Do I need MCP?
No. MCP makes the connection shorter, but every route marked "yes" in the table works over the REST API today.
Is the Outreach2day API included in my plan?
Yes. API access is part of every plan; you pay for mailboxes and domains. See pricing.
Is there an Outreach2day MCP server?
A hosted one is in early access. You can also build a local one over the public API in about 110 lines of Python.
Related
Outreach2day MCP: the hosted MCP server and its per-client pages (early access)
Cold email API: endpoints, limits and errors
Cold email MCP server: the local server the guides use
Create an account, add mailboxes, then create a key for your assistant.
Keep reading
Send 100,000 cold emails a month
How to send 100,000 cold emails a month while keeping quality, engagement and conversion rates high: domains, mailboxes, warm-up, leads, copy and scaling.
Step-by-step playbook · 19 min readEmail tools5 min read
How to Connect ChatGPT to Cold Email
Email tools5 min read
How to Connect Claude Code to Cold Email
Email tools5 min read
How to Connect Claude to Cold Email
