February 25, 2026

Build an AI Sales Prospecting Pipeline with MCP Tools

A manual SDR spends roughly 40% of their time on prospecting — finding contacts, verifying emails, researching companies, writing personalized outreach. For a team of five SDRs billing $70K each, that's $140K/year spent on work that follows the same process every time with different inputs.

An ai sales pipeline built on MCP tools can automate 80-90% of that workflow. Not the judgment calls — who to target, what messaging angle to use, whether a reply signals real interest — but the mechanical work of finding, verifying, researching, and drafting.

This guide walks through building a complete ai agent lead generation workflow using four MCP servers that chain together into a single pipeline. Each step builds on the previous one, and the full config is at the end so you can copy and run it.

What an AI Sales Prospecting Pipeline Looks Like

The pipeline has five stages, each handled by a different MCP tool:

Target Definition (you)
    ↓
Prospect Discovery (Apollo + Tavily)
    ↓
Email Verification (Reoon)
    ↓
Campaign Creation (Instantly)
    ↓
Reply Management (AgentMail)

You define the target — "VP-level engineering leaders at Series B fintech companies in Boston." The agent handles everything else: finding matching contacts, verifying their email addresses, researching their companies for personalization, drafting outreach, and managing the sending.

The critical design principle: each stage passes structured data to the next. Apollo outputs contacts with emails. Reoon filters to verified-only. Tavily adds company research. Instantly sends the campaigns. AgentMail handles replies. No manual data entry between stages.

This maps to the prompt chaining pattern — sequential steps where each consumes the prior output, with programmatic gates between them. The verification step acts as a quality gate: only verified contacts proceed to outreach.

Step 1: Find Prospects with Apollo and Tavily

The discovery stage combines two tools: Apollo.io for contact data and Tavily for company research.

Apollo for contact search

Apollo is a B2B contact database with search filters for role, seniority, company size, industry, funding stage, location, and more. The MCP server exposes these filters to your agent.

A typical prospecting prompt:

Search Apollo for:
- Title: VP of Engineering, CTO, Head of Engineering
- Company size: 50-500 employees
- Funding: Series B
- Industry: Fintech
- Location: Boston, MA
- Limit: 25 contacts

Return name, title, company, email, and LinkedIn URL for each.

The agent calls Apollo's search endpoint and returns structured contact data. Each result includes the company domain, which feeds into the next step.

Tavily for company research

Raw contact data isn't enough for good outreach. You need context — what the company does, recent news, product launches, hiring signals. Tavily fills this gap:

For each of these 25 companies, do a brief Tavily search to find:
- What the company does (one sentence)
- Any recent news from the past 3 months
- Current job openings in engineering (hiring signal)

Claude Code's parallel execution shines here — it can research all 25 companies simultaneously rather than sequentially. What would take 25 serial searches completes in roughly the time of one.

The output: a list of contacts enriched with company context. This enrichment is what makes the difference between "Hi [Name], I noticed your company is doing interesting work" (garbage) and "Hi Sarah, I saw Acme just raised their Series B and is hiring three backend engineers — we help scaling eng teams ship 30% faster" (relevant).

For more on setting up Tavily, see our Tavily MCP setup guide.

Step 2: Verify Emails with Reoon

This is the stage most teams skip and then regret. Sending to unverified email addresses damages your sender reputation, increases bounce rates, and can get your sending domain blacklisted. At scale, a 15% bounce rate from stale data can tank your deliverability for weeks.

Reoon verifies each email address before it enters the outreach pipeline:

Verify all 25 email addresses through Reoon. 
Remove any that come back as invalid or high-risk.
Report how many passed and how many were filtered.

The agent sends each address through Reoon's verification API, which checks:

Syntax — Is the email format valid?
Domain — Does the domain exist and accept email?
Mailbox — Does the specific mailbox exist?
Risk scoring — Is this a catch-all, disposable, or role-based address?

Typical results: 20-22 out of 25 pass verification. The 3-5 that fail are exactly the ones that would have bounced and hurt your reputation. Cheap insurance.

Setting the quality threshold

You can tune the filter:

Keep only addresses that Reoon rates as "valid" or "accept-all with low risk."
Remove anything rated "invalid," "unknown," or "high-risk."

For cold outreach, be conservative. Sending to 20 verified addresses outperforms sending to 25 where 5 bounce.

Step 3: Launch Outreach with Instantly

With verified contacts and company research in hand, the agent creates a campaign in Instantly.ai.

Draft personalization

First, the agent generates personalized email copy for each prospect using the company research from Step 1:

For each verified contact, draft a personalized cold email that:
- References something specific about their company (from the research)
- Connects it to our value proposition: [your product description]
- Keeps the email under 120 words
- Includes a clear, low-commitment CTA (15-min call, not "buy our product")
- Uses their first name, not full name

Show me all drafts before sending anything.

The "show me all drafts" instruction is your approval gate. Never let an agent send outreach without human review, especially during initial setup. Read every email, check that the personalization is accurate and relevant, and only approve once you're satisfied.

Campaign creation

After approval:

Create an Instantly campaign called "Fintech VPs Boston - Feb 2026"
with the approved emails. Settings:
- Send 5 emails per day
- Start on Monday
- Follow-up sequence: Day 3 bump, Day 7 new angle
- Warm-up: enabled

The agent creates the campaign in Instantly with the specified parameters. Instantly handles the sending infrastructure — warm-up, sender rotation, deliverability monitoring, and scheduling.

Follow-up sequences

Outreach rarely converts on the first email. The agent can draft follow-up sequences:

For each prospect, draft two follow-ups:
1. Day 3: Brief bump referencing the original email (2-3 sentences)
2. Day 7: New angle that adds value — share a relevant insight or resource

These get added to the Instantly campaign as automated follow-ups.

Handling Replies and Follow-Ups with AgentMail

Sending is half the pipeline. The other half is handling what comes back.

Instantly handles campaign sends, but when prospects reply, you need conversational email management. That's where AgentMail comes in — it's built for AI agents that need to participate in back-and-forth email threads.

Positive replies

When a prospect expresses interest:

Monitor for positive replies. When someone responds with interest:
- Acknowledge their reply within 2 hours
- Suggest 3 specific time slots for a call next week
- Include a brief agenda for the call
- CC the account executive

Objections

When a prospect pushes back:

For objection replies (not interested, bad timing, wrong person):
- "Not interested": Thank them, ask if they'd prefer to be removed
- "Bad timing": Acknowledge, ask when would be better, set a reminder
- "Wrong person": Ask who the right contact would be, thank them

Out-of-office

For out-of-office replies: Note the return date and reschedule 
the follow-up sequence to resume 2 days after they return.

AgentMail handles the email reading, parsing, and response generation. The agent classifies each reply and follows the appropriate playbook. Human oversight remains critical here — review the agent's reply classifications weekly and correct any miscategorizations.

Full Pipeline Config and Optimization Tips

Complete MCP configuration

Here's the full config for Claude Code with all four pipeline tools:

{
  "mcpServers": {
    "tavily": {
      "command": "npx",
      "args": ["-y", "tavily-mcp@latest"],
      "env": { "TAVILY_API_KEY": "${TAVILY_API_KEY}" }
    },
    "apollo": {
      "command": "npx",
      "args": ["-y", "@apollo/mcp-server"],
      "env": { "APOLLO_API_KEY": "${APOLLO_API_KEY}" }
    },
    "reoon": {
      "command": "npx",
      "args": ["-y", "@reoon/mcp-server"],
      "env": { "REOON_API_KEY": "${REOON_API_KEY}" }
    },
    "instantly": {
      "command": "npx",
      "args": ["-y", "@instantly/mcp-server"],
      "env": { "INSTANTLY_API_KEY": "${INSTANTLY_API_KEY}" }
    },
    "agentmail": {
      "command": "npx",
      "args": ["-y", "@agentmail/mcp-server"],
      "env": { "AGENTMAIL_API_KEY": "${AGENTMAIL_API_KEY}" }
    }
  }
}

For OpenClaw, the equivalent YAML:

mcp_servers:
  tavily:
    command: npx
    args: ["-y", "tavily-mcp@latest"]
    env:
      TAVILY_API_KEY: "${TAVILY_API_KEY}"
  apollo:
    command: npx
    args: ["-y", "@apollo/mcp-server"]
    env:
      APOLLO_API_KEY: "${APOLLO_API_KEY}"
  reoon:
    command: npx
    args: ["-y", "@reoon/mcp-server"]
    env:
      REOON_API_KEY: "${REOON_API_KEY}"
  instantly:
    command: npx
    args: ["-y", "@instantly/mcp-server"]
    env:
      INSTANTLY_API_KEY: "${INSTANTLY_API_KEY}"
  agentmail:
    command: npx
    args: ["-y", "@agentmail/mcp-server"]
    env:
      AGENTMAIL_API_KEY: "${AGENTMAIL_API_KEY}"

Optimization tips

Start small. Run the pipeline with 10 contacts first. Review every output. Once the pattern is proven, scale to 25, then 50. Don't automate what you haven't validated.

Track metrics at each stage. How many contacts does Apollo return? What's the Reoon pass rate? What's the open rate in Instantly? What's the reply rate? Each stage has a conversion rate, and knowing them tells you where to optimize.

Separate targeting from execution. Define your ICP (ideal customer profile) in a file. When you want to run the pipeline for a different segment, change the targeting file — not the pipeline instructions. This makes the workflow reusable across market segments.

Review weekly, not daily. Once the pipeline is running, do a weekly audit: review reply classifications, check for false positives/negatives in verification, and update personalization templates based on what's getting responses.

Mind the reliability math. Five tools chained together with 95% per-step reliability gives you 77% overall. Add error handling at each stage — if Reoon times out, retry with backoff. If Apollo returns no results, broaden the search criteria automatically. For production-grade patterns, see our agent automation guide.

Frequently Asked Questions

What MCP tools do I need for an AI sales prospecting pipeline?

A complete ai sdr pipeline setup uses five MCP servers: Apollo.io for finding contacts (search by role, company, industry, funding), Tavily for company research and personalization context, Reoon for email verification, Instantly.ai for campaign sending with warm-up and deliverability management, and AgentMail for handling replies conversationally. You can start with just Apollo + Reoon + Instantly for a minimal pipeline, then add Tavily for better personalization and AgentMail for reply management as the workflow matures.

How many prospects should I run through the pipeline at once?

Start with 10 contacts to validate the full pipeline — review every output at every stage. Once proven, scale to 25-50 per run. Most teams send 5-10 emails per day per sending account to maintain deliverability, so a batch of 50 verified contacts with follow-up sequences covers roughly two weeks of sending. Running the pipeline weekly with fresh prospects keeps a steady flow without overwhelming your sending infrastructure.

How much does an AI sales prospecting pipeline cost?

Tool costs vary: Apollo has a free tier with limited searches and paid plans from $49/month. Reoon offers trial verifications and paid plans based on volume. Instantly starts at $30/month with built-in warm-up. Tavily's free tier (1,000 searches/month) covers most prospecting research needs. AgentMail has free tier limits. LLM costs for the agent itself depend on volume — a typical 50-prospect pipeline run uses 200,000-400,000 tokens. Total monthly cost for a small team: $100-300 in tools plus $50-150 in LLM costs, replacing $5,000-10,000/month in SDR prospecting time.

Should I let the AI agent send emails without review?

No — at least not initially. Always include a human approval gate before sending outreach, especially during the first few weeks. Review every email for accurate personalization, appropriate tone, and correct company references. Once you've reviewed 100+ emails and the quality is consistently good, you can relax to spot-checking a random sample from each batch. Never fully remove human oversight from cold outreach — the reputational risk of a badly personalized email isn't worth the time saved.

Can I use this pipeline with OpenClaw instead of Claude Code?

Yes. The MCP tools are identical — same servers, same config format (YAML instead of JSON), same capabilities. OpenClaw's skill system adds an additional layer: you can package the pipeline instructions into a skill file that the agent reads automatically when triggered. This makes the workflow more structured and repeatable than ad-hoc Claude Code sessions. See our OpenClaw workflow tutorial for the skill-based approach to building pipelines.