How to Build an AI SDR Pipeline with Claude Code and Winnr

What Is an AI SDR?

An AI SDR (AI sales development representative) is an AI-powered system that handles the operational side of outbound sales. But let's be specific about what that means — and what it doesn't.

An AI SDR manages infrastructure:

• Domain provisioning (search, purchase, DNS configuration)
• Mailbox creation and credential management
• Email warming (ramp-up, health monitoring, pause/resume)
• Deliverability monitoring (inbox rates, spam rates, DNS health)
• Sequencer integration (credential export in the right format)
• Troubleshooting (diagnosing and resolving deliverability issues)

An AI SDR does not handle:

• Email copywriting (that's your sequencer or a separate AI tool)
• Lead generation (that's Apollo, Clay, or your own research)
• Reply management (that's your sequencer's inbox or CRM)

This separation is intentional. Infrastructure management is operationally complex — dozens of domains, hundreds of mailboxes, DNS records, warming schedules, health scores — but it follows deterministic patterns. That makes it ideal for AI automation. The creative work (copy, targeting, relationship building) stays with humans or purpose-built tools.

Why 2026 Is the Inflection Point

The idea of AI-managed outbound infrastructure isn't new. What's new is that three trends converged in 2025-2026 to make it practical:

1. MCP standardized AI tool use

Anthropic's Model Context Protocol created a universal interface for AI assistants to call external tools. Before MCP, every integration was bespoke. Now Claude, Cursor, Windsurf, and other AI assistants can connect to any MCP-compatible service with a single config block. Winnr was the first cold email platform to ship an MCP server.

2. Cold email infrastructure APIs matured

Programmatic domain purchase, mailbox provisioning, DNS management, and warming control became available through REST APIs. Winnr's API covers the full lifecycle — 37 endpoints from domain search to credential export — which means an AI agent has everything it needs to manage infrastructure without human intervention.

3. Claude Code skills enable domain-specific workflows

Claude Code skills are reusable, domain-specific prompts that encode best practices into slash commands. Instead of telling the AI "search for domains, check availability, buy the cheapest ones, set up DNS, create mailboxes, enable warming with a slow ramp" every time, you run /winnr setup and the skill handles the entire workflow with proven defaults. Skills turn tribal knowledge into executable automation.

The result: for the first time, AI can manage the full infrastructure lifecycle — from zero domains to sending-ready mailboxes — without writing code, clicking dashboards, or remembering the 47 steps involved.

The Architecture

An AI SDR pipeline has three layers. Each layer is independent, composable, and replaceable:

+--------------------------------------------------+
|              CAMPAIGN LAYER                       |
|  Sequencer (Smartlead, Instantly, Lemlist, etc.)  |
|  Lead lists, email copy, send scheduling, replies |
+--------------------------------------------------+
                      |
             credential export
                      |
+--------------------------------------------------+
|            INTELLIGENCE LAYER                     |
|  Claude Code + Winnr Skills + MCP Server          |
|  Orchestration, best practices, monitoring        |
|                                                   |
|  /winnr setup    - provision infrastructure       |
|  /winnr health   - monitor deliverability         |
|  /winnr export   - export to sequencer            |
|  /winnr scale    - add capacity                   |
|  /winnr troubleshoot - diagnose issues            |
+--------------------------------------------------+
                      |
                  REST API
                      |
+--------------------------------------------------+
|           INFRASTRUCTURE LAYER                    |
|  Winnr (domains, mailboxes, DNS, warming)         |
|  Managed SMTP/IMAP, deliverability engine         |
+--------------------------------------------------+

Infrastructure layer (Winnr) handles the hard parts: domain registration, DNS record management, mailbox provisioning on dedicated mail servers, DKIM signing, SPF/DMARC configuration, and email warming with real engagement signals.

Intelligence layer (Claude Code) orchestrates operations through natural language. The MCP server provides raw API access (37 tools). Skills add domain expertise on top — encoding warming best practices, health thresholds, scaling strategies, and troubleshooting playbooks into reusable commands.

Campaign layer (your sequencer) handles what infrastructure can't: lead lists, email copy, send scheduling, A/B testing, and reply management. Winnr exports credentials in 15+ sequencer formats, so this layer plugs in without custom integration.

Step-by-Step: From Zero to Sending

Here's the full tutorial. Total active time: about 5 minutes. Total wall-clock time: 14-21 days (warming).

Step 1: Set Up the Stack (5 minutes)

You need three things: a Winnr account, the MCP server, and Claude Code skills.

Create your Winnr account and API token:

1. Sign up at app.winnr.app
2. Go to Settings → API Tokens
3. Create a read+write token (starts with wnr_)

Add the Winnr MCP server to Claude Code:

claude mcp add winnr -- env WINNR_API_TOKEN=wnr_your_token_here uvx winnr-mcp

That's one command. Claude Code now has access to all 37 Winnr tools — domain search, mailbox creation, warming control, inbox access, and more.

Install the Winnr skills pack:

curl -sL https://winnr.app/skills/install.sh | bash

This installs five slash commands: /winnr setup, /winnr health, /winnr export, /winnr scale, and /winnr troubleshoot. Each skill encodes best practices so you don't have to remember optimal warming ramp rates or DNS verification sequences.

Step 2: Provision Infrastructure (2 minutes)

Run the setup skill:

/winnr setup

The skill will ask you a few questions (how many domains, mailboxes per domain, your industry/niche for domain name suggestions) and then execute the full provisioning workflow:

1. Domain search: Generates domain name variations based on your niche, checks bulk availability, filters by price
2. Domain purchase: Buys the best available domains (charges your card on file via Stripe)
3. DNS configuration: Sets up MX, SPF, DKIM, and DMARC records automatically, then verifies propagation
4. Mailbox creation: Creates professional mailboxes (first.last@domain.com) with strong generated passwords
5. Warming activation: Enables warming on every mailbox with an appropriate ramp-up schedule

A typical setup — 10 domains, 4 mailboxes each, 40 mailboxes total — takes about 2 minutes of wall-clock time. The skill handles the sequencing, waits for async jobs to complete, and reports the final status.

You: /winnr setup

Claude: I'll help you set up your cold email infrastructure. A few questions:

  1. How many domains do you want? (recommended: 5-20)
  2. Mailboxes per domain? (recommended: 3-5)
  3. What's your industry/niche? (for domain name suggestions)

You: 10 domains, 4 mailboxes each, B2B SaaS sales consulting

Claude: Setting up 10 domains with 4 mailboxes each (40 total)...

  [1/5] Searching for domains... found 47 available .com domains
  [2/5] Purchasing top 10 by relevance... 10/10 purchased
  [3/5] Configuring DNS (MX, SPF, DKIM, DMARC)... 10/10 verified
  [4/5] Creating 40 mailboxes... 40/40 created
  [5/5] Enabling warming (slow ramp-up)... 40/40 warming

  Done. 10 domains, 40 mailboxes, all warming.
  Estimated sending readiness: April 12-19, 2026.
  Run /winnr health daily to monitor progress.

Step 3: Wait for Warming (14-21 days)

Warming is the one step that can't be rushed. New mailboxes need to build sender reputation by exchanging real emails with real inboxes over 2-3 weeks. During this period, run the health check daily:

/winnr health

The skill queries warming metrics for all your mailboxes and produces a summary report:

Healthy report (day 14):

Winnr Health Report — March 29, 2026

  40 mailboxes across 10 domains
  38 healthy (health score 75+, inbox rate 90%+)
   2 warming (health score 60-74, on track)
   0 at risk

  Avg health score: 82 | Avg inbox rate: 93%
  Recommendation: On track. 3-5 more days until full readiness.

Concerning report (needs attention):

Winnr Health Report — March 29, 2026

  40 mailboxes across 10 domains
  35 healthy (health score 75+, inbox rate 90%+)
   3 warming (health score 60-74, on track)
   2 at risk (health score below 50)

  At-risk mailboxes:
    sarah.jones@growthpeak.com — health 42, inbox 67%, spam 18%
    mike.chen@salessurge.com — health 38, inbox 61%, spam 22%

  Recommendation: Run /winnr troubleshoot for the 2 at-risk mailboxes.
  Consider pausing warming and investigating DNS/content issues.

Step 4: Export to Your Sequencer (30 seconds)

Once mailboxes are warm, export credentials to your sequencer:

/winnr export smartlead

The skill performs a readiness check first — it won't export mailboxes that aren't ready — then generates a CSV in the exact format your sequencer expects:

You: /winnr export smartlead

Claude: Running readiness check...

  38/40 mailboxes ready (health 75+, inbox rate 90%+)
   2 mailboxes excluded (still warming)

  Exported 38 mailboxes to smartlead_export_2026-03-29.csv
  Columns: email, password, imap_host, imap_port, smtp_host, smtp_port, name

  Import this CSV into Smartlead → Settings → Email Accounts → Bulk Import.

Winnr supports 15+ export formats: Smartlead, Instantly, Snov, Lemlist, Woodpecker, Reply, Mailshake, GMass, Yesware, Mixmax, Outreach, Salesloft, and more. Same command, different format name.

Step 5: Monitor and Maintain (Ongoing)

Once you're sending, check health weekly:

• /winnr health — weekly deliverability report across all mailboxes
• /winnr troubleshoot — diagnose and resolve issues when health drops
• /winnr scale 20 — add more domains and mailboxes when you need capacity

The troubleshoot skill checks DNS records, warming metrics, sending patterns, and inbox/spam rates to identify the root cause. It then recommends specific actions: pause a mailbox, adjust warming volume, fix a broken DKIM record, or rotate a burned domain.

Advanced: Combining with Other Tools

The AI SDR pipeline handles infrastructure. Here's how it fits into a full outbound stack:

Lead generation

Use Apollo, Clay, Ocean.io, or your own scraping to build target lists. These feed into your sequencer, not into Winnr. The infrastructure layer doesn't care who you're emailing — it just makes sure your emails land in inboxes.

AI copywriting

Use Claude, GPT, or a specialized tool (Lavender, Regie.ai) to write personalized email copy. Again, this feeds into your sequencer. Winnr handles the sending infrastructure, not the content.

CRM integration

Your sequencer handles reply detection and CRM sync (Salesforce, HubSpot, Pipedrive). Winnr's inbox tools can also read replies programmatically if you want to build custom reply routing with AI.

The composability principle

Each layer is replaceable. Don't like Smartlead? Switch to Instantly — just re-export credentials. Want to use a different lead source? Swap Apollo for Clay. The infrastructure layer (Winnr + Claude Code) stays constant regardless of what you plug in above and below it.

ROI Analysis: Manual vs. AI-Managed

Here's the math on a typical setup: 10 domains, 40 mailboxes.

At $50/hour ops cost:

• Manual setup: $200-400 per setup cycle
• AI-managed setup: ~$4 (5 minutes)
• Savings per cycle: $196-396

If you set up new infrastructure monthly (common for agencies and high-volume senders), that's $2,400-4,800/year in ops time saved — before counting the avoided mistakes.

The less quantifiable benefit: consistency. The AI follows the same best practices every time. It never forgets to set up DMARC, never misconfigures a DKIM record, never skips the warming ramp-up period. Human ops teams make these mistakes regularly, and each one costs days of deliverability recovery.

Frequently Asked Questions

Can the AI SDR write my cold emails?

No — and that's by design. The AI SDR manages infrastructure: domains, mailboxes, warming, deliverability, and sequencer integration. Email copywriting, lead targeting, and send scheduling are separate concerns handled by your sequencer and copywriting tools. This separation keeps each layer focused and replaceable.

How much does this cost?

Three cost components: Winnr plan ($49-249/month depending on domain and mailbox count), domains ($8-15/domain/year), and Claude Code (Anthropic API key or Claude Max subscription). The MCP server is open source and free. Skills install with one command at no cost. For a typical 10-domain, 40-mailbox setup, expect ~$100-150/month all-in for infrastructure.

What sequencers work with this?

Winnr exports credentials in 15+ formats: Smartlead, Instantly, Snov, Lemlist, Woodpecker, Reply, Mailshake, GMass, Yesware, Mixmax, Outreach, Salesloft, and generic IMAP/SMTP CSV. If your sequencer can import email accounts via CSV, it works.

Is this reliable enough for production?

Yes. All operations go through Winnr's REST API with rate limiting (300-500 req/min), authentication, error handling, and account scoping. The MCP server is a thin HTTP client — it has no direct infrastructure access. Skills encode proven best practices (warming ramp rates, health thresholds, DNS verification sequences) so the AI follows playbooks rather than improvising. The MCP server is open source, so you can inspect every line.

What is an AI SDR and how is it different from a human SDR?

An AI SDR handles the operational side of sales development: infrastructure setup, warming, monitoring, and maintenance. A human SDR handles the relational side: personalized outreach, relationship building, objection handling, and deal progression. The AI SDR is the ops layer underneath your outbound strategy — it makes sure your emails reach inboxes so your human SDRs (or AI copywriters) can focus on what they say, not whether it gets delivered.

Start Building Your AI SDR Pipeline

Everything you need is available today:

• Sign up for Winnr — create an account and API token (free tier available)
• Install Claude Code skills — one curl command, five slash commands
• MCP server docs — full tool reference, setup guides, security details
• GitHub repository — open source MCP server code
• API documentation — full REST API reference

The infrastructure layer is the hardest part of outbound to get right and the easiest to automate. Let the AI handle the ops so you can focus on what actually closes deals.

Related Reading

• Winnr MCP Server: Let AI Agents Manage Your Cold Email Infrastructure — deep dive into the MCP server's 37 tools and security model
• How to Embed Email Infrastructure in Your App with the Winnr API — developer guide for direct REST API integration
• Vibe-Code Your Own Email Sequencer — build a custom outreach tool using Winnr's API
• Getting Started with Winnr — 5-minute quickstart for domains, mailboxes, and warming
• Cold Email Best Practices — deliverability guide covering sending limits, authentication, and content