Add https://mcp.winnr.app/mcp as a custom connector (Claude) or MCP server (ChatGPT, Cursor, VS Code), then sign in with Winnr and choose what the assistant may do. Nothing to install and no API key. The assistant then sees 55 tools for managing domains, mailboxes, warming and the inbox. A local version (uvx winnr-mcp with an API token) is also available for desktop apps.
Two ways to connect
Hosted (recommended). Add https://mcp.winnr.app/mcp as a connector, sign in with Winnr, choose what the assistant may do. Works with claude.ai on web and mobile, ChatGPT, and every desktop client. Nothing to install, no API key to keep safe, and you can revoke it from the API page at any time.
Local. Run the open-source package on your own machine with an API token: uvx winnr-mcp. Desktop apps only. Use it if you want the server in your own environment.
What "MCP" actually is
Model Context Protocol is a way for AI assistants to call external tools. Instead of you writing code that calls Winnr's API, the AI does it directly — you just describe what you want.
Example conversation with Claude, once MCP is set up:
You: Provision 3 new domains for a Q4 outreach campaign. Use .com TLDs. Then create 5 mailboxes on each with first names Sam, Alex, Jordan, Taylor, and Morgan. Enable warming on all of them.
Claude: [checks prices with winnr_search_domains_bulk, asks you to confirm the total, calls winnr_purchase_domains once, polls the job, then winnr_bulk_create_email_users per domain and winnr_enable_warming across all 15 mailboxes]
That whole workflow takes about 60 seconds with MCP. Without MCP, you're either clicking through the UI or writing custom scripts.
When MCP is worth setting up
- AI SDR pipelines. Let an agent handle infrastructure provisioning end-to-end.
- Ops automation. Ask Claude to "audit my domain health" and get a full report without leaving the chat.
- One-off complex tasks. Bulk operations that don't fit the UI's bulk actions.
Security notes
- The MCP server holds an API token with your account's full permissions. Treat it like you would any privileged credential.
- Read/write scope is required for most tools. Use read-only if you only want the AI to read (safe for exploratory chats) — the server then hides every tool that creates, sends, buys or deletes.
- Only four tools charge your card (buying domains, buying pre-warmed domains, enabling warming). The server instructs the assistant to state the exact items and price and get your explicit yes first.
- Revoke the token any time from the API page if you suspect it's leaked.
What's next
- Set it up in the app — guided, token pre-filled, connection check
- Create an API token
- MCP GitHub repo
- Winnr MCP marketing page
Step-by-step
-
1. Open the AI Assistants page in the app
Log in and go to app.winnr.app/mcp. Pick your assistant and follow the steps it shows. The page turns green once your assistant makes its first call.
-
2. Add the connector (hosted, recommended)
In Claude (web, mobile or desktop) go to Settings, Connectors, Add custom connector, and paste https://mcp.winnr.app/mcp. In ChatGPT use Settings, Apps & Connectors (called Connectors in older builds), Advanced settings, Developer mode, then Create. In Cursor, Windsurf or VS Code add an MCP server with that URL. Then click Connect.
-
3. Sign in and choose access
Your browser opens Winnr. Pick what the assistant may do — see only, make changes, or also spend money. Approve, and you are returned to the assistant.
-
4. Create a Winnr API token (local install only)
API → Create Token with read/write scope (or read-only if the assistant should only report). Copy the token (starts with wnr_). Only needed for the local uvx install.
-
5. Add to Claude Desktop config
Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%/Claude/claude_desktop_config.json (Windows). Add — {"mcpServers": {"winnr": {"command": "uvx", "args": ["winnr-mcp"], "env": {"WINNR_API_TOKEN": "wnr_YOUR_TOKEN"}}}}
-
6. Restart Claude Desktop
Fully quit and reopen. The MCP server auto-connects on startup. You'll see the tools appear in Claude's tool list.
-
7. For Cursor
Add to ~/.cursor/mcp.json — {"mcpServers": {"winnr": {"command": "uvx", "args": ["winnr-mcp"], "env": {"WINNR_API_TOKEN": "wnr_YOUR_TOKEN"}}}}. Restart Cursor.
-
8. For Claude Code
Run: claude mcp add --scope user winnr -e WINNR_API_TOKEN=wnr_YOUR_TOKEN -- uvx winnr-mcp. Then /mcp inside Claude Code shows Winnr connected.
-
9. Test it
Ask the AI — 'List my Winnr domains.' It should call the winnr_list_domains tool and return your domain list. If it doesn't see the tool, restart the AI host and check WINNR_API_TOKEN is set correctly.
Frequently asked questions
Do I need to install anything?
No. The hosted server at https://mcp.winnr.app/mcp is added as a connector and you sign in with your Winnr login. A local install (uvx winnr-mcp with an API token) is available for desktop apps if you prefer to run it yourself.
Does it work with ChatGPT and the Claude mobile app?
Yes, through the hosted server. Anything that supports remote MCP connectors can use it — claude.ai on web and mobile, ChatGPT developer-mode connectors, Claude Desktop, Cursor, Windsurf and VS Code.
Can the AI spend my money without asking?
No. Buying domains, buying pre-warmed domains and enabling warming are two-step. The first call only returns a quote with exact prices, and nothing is charged until the assistant calls again with your confirmation. You can also grant access without the spend permission at all, and the tools that charge are then hidden entirely.
How do I disconnect an assistant?
Go to the API page in Winnr and revoke the token named 'MCP ·
What can the MCP server do?
55 tools — provision domains, create mailboxes, enable/tune warming, read metrics, export credentials, purchase marketplace domains, read the inbox, tag and organize. Full CRUD across every Winnr resource.
What AI hosts is it compatible with?
Anything that speaks the Model Context Protocol. Hosted — claude.ai web and mobile, ChatGPT, Claude Desktop, Cursor, Windsurf, VS Code. Local — any desktop client that runs stdio servers.
Is the MCP server open source?
Yes — the source is on GitHub and the package is published on PyPI as winnr-mcp.
Do I need to pay extra for MCP?
No. MCP calls Winnr's regular API under the hood, so it uses your account's normal rate limit and doesn't add cost. The MCP server itself is free.
Can I self-host the MCP server?
Yes, it's a plain Python package. Run it however you want. Most people use uvx for zero-install convenience.