Claude Code Lead Agent: Build an AI Prospecting Agent in 2026

TL;DR

• A Claude Code lead agent scrapes target companies, enriches contacts via SyncGTM, scores them against your ICP, and pushes qualified records to your CRM — all from a terminal prompt.
• No production code required. Claude Code writes and runs the scripts on your behalf.
• Five steps: scraping → enrichment → scoring → routing → pipeline wiring.
• Typical enrichment coverage: 80%+ on a cold ICP list when using SyncGTM waterfall across 20+ providers.
• Total cost at 500 leads/week: roughly $15–40 in Claude API + SyncGTM credits. Far cheaper than Clay or Apollo at scale.

What Is a Claude Code Lead Agent?

A Claude Code lead agent is an autonomous prospecting workflow built on Anthropic's Claude Code platform that executes scraping, enrichment, scoring, and CRM routing — without a human touching each step.

You describe what you want in plain language: your ICP, your scoring criteria, your CRM destination. Claude Code writes the scripts, calls the APIs, and delivers qualified, enriched leads ready for outreach.

No UI to learn. No workflow builder to configure. No seat license. The agent runs on your machine, hits your API keys, and follows your rules.

According to Gartner, over 60% of B2B sales teams will use AI agents for pipeline tasks by 2027. Building your own gives you data control and cost advantages that hosted tools cannot match.

Why Build Instead of Buy

Apollo charges ~$49–99/mo per user. Clay's enrichment credits add up fast. Every hosted tool adds a subscription and a data silo you don't control.

A Claude Code lead agent changes the economics: you pay only for Claude API tokens and SyncGTM enrichment credits. At 500 leads/week, that's $15–40 — versus $200+ on hosted platforms.

The control matters more than the cost. Your ICP logic lives in a prompt you edit in 30 seconds. Your scoring rules are explicit, not a black-box algorithm. Your CRM routing matches your exact field structure — no remapping.

The Claude Code sales agent guide covers the outbound half of this workflow (email personalization, follow-up sequences). This post handles the top-of-funnel half: finding and qualifying leads before outreach begins.

Prerequisites

Before starting, confirm you have:

• Claude Code installed — available via npm install -g @anthropic/claude-code. Requires Node.js 18+.
• Anthropic API key — set as ANTHROPIC_API_KEY in your environment.
• SyncGTM API key — for waterfall enrichment. Get one at syncgtm.com/pricing.
• CRM MCP server — for HubSpot or Salesforce, use Composio's managed MCP. For Close or Pipedrive, use their native MCP servers.
• Target list source — a LinkedIn Sales Navigator URL, a TheirStack company search, or a CSV of company domains. The agent needs somewhere to start scraping.

You do not need Python, a cloud server, or production engineering experience. Claude Code handles execution. You handle intent.

Step 1: Automated Lead Scraping

The agent starts by building a raw company list — finding ICP-matched targets and extracting the domains you need for enrichment.

Give Claude Code your target criteria in a single prompt. Example:

Find SaaS companies in the US with 50–500 employees using HubSpot.
Export: company name, domain, LinkedIn URL, estimated headcount, industry.
Use TheirStack to filter by HubSpot tech stack.
Save results to leads-raw.csv.

Claude Code writes a Python script calling the TheirStack API, handles pagination, deduplicates results, and saves the output — no script writing on your end.

For LinkedIn-based scraping, Claude Code drives a Playwright browser session to export Sales Navigator results. That approach is covered in the Claude Code sales contact guide.

Aim for 200–500 companies per run. Larger batches take longer to enrich and harder to review — run daily on a fresh ICP slice rather than weekly on a stale dump.

Step 2: SyncGTM Enrichment

Raw domains don't close deals. You need verified work email, direct dial, job title, seniority, and firmographic context before outreach can start.

Tell Claude Code to run SyncGTM's waterfall enrichment:

Enrich leads-raw.csv using the SyncGTM API.
For each company, find: VP of Sales, VP of Marketing, or Head of Revenue.
Return: full name, work email, direct dial, LinkedIn URL, job title, seniority.
Waterfall across all available providers.
Save to leads-enriched.csv.

SyncGTM's waterfall runs each domain through 20+ providers in priority order — if Apollo misses, Hunter fills it; if Hunter misses, Prospeo fills it. That cascading coverage drives email hit rates above 80%, versus 40–55% from a single provider.

The Claude Code lead enrichment guide documents every enrichment action in the SyncGTM API, including advanced patterns like tech stack lookups and buyer intent signals.

After enrichment, the CSV holds everything the scoring step needs: verified contact identity, firmographics, and confirmed contact channels.

Step 3: Lead Scoring Without a Data Scientist

Most teams stall here. Building a predictive scoring model requires historical CRM data, labeled examples, and data engineering. Claude Code skips the model entirely — and scores just as effectively for most ICPs.

Instead of a trained classifier, you define explicit point rules. Example prompt:

Score each contact in leads-enriched.csv from 0–100.
Award points:
  +30 if headcount is 51–500
  +20 if industry is SaaS, Fintech, or HR Tech
  +20 if seniority is VP or above
  +15 if company uses HubSpot (tech stack match)
  +15 if job title includes "Revenue", "Sales", or "Growth"
Deduct:
  -50 if headcount is under 10 or over 2000
  -30 if industry is non-profit or government
Save score to leads-scored.csv. Sort descending.

Claude Code writes and runs the scoring script, returning a ranked list with a numeric score per contact. You wrote a business rule — not a line of code.

Rule-based scoring is an advantage, not a compromise. Every point has an explicit reason you can explain to any SDR or manager — and you can tune the logic in 60 seconds when your ICP shifts.

Predictive scoring tools like MadKudu cost $500–2,000/mo for mid-market teams. This approach costs $0 in tooling — just Claude API tokens.

Set routing thresholds: 60+ routes to CRM, 30–59 goes to a manual review file, below 30 is discarded. Adjust after your first week of data.

Step 4: CRM Routing and Assignment

Scored leads need to land in the right place. Claude Code connects to your CRM via MCP and creates or updates records based on your routing rules — no manual import.

Example prompt for HubSpot routing:

Push contacts from leads-scored.csv where score >= 60 to HubSpot.
Create a new contact if the email doesn't already exist.
Update existing contacts if the email matches.
Set fields: First Name, Last Name, Email, Phone, Job Title, Company, Lead Score.
Set Lead Status to "New" and assign to the owner with the fewest open leads.
Add tag: "claude-agent-[today's date]".

The MCP layer handles OAuth, rate limits, and field mapping automatically. Claude Code sends plain-language instructions; the MCP server translates them to HubSpot API calls and returns a confirmation of records created and updated.

For Salesforce, the pattern is identical — swap in the Salesforce MCP server. The Claude Code CRM integration guide covers both platforms and custom MCP setup for any CRM with a REST API.

Always add a deduplication check before each push. Claude Code queries the CRM for existing contacts by email first — preventing duplicate records from stacking up across daily runs.

Step 5: Wire the Full Pipeline

Steps 1–4 run as individual prompts while you validate each stage. Once each step works, wire them into a single skill file that runs end-to-end.

Create a Claude Code skill file that defines the full workflow:

# lead-agent.md

## Inputs
- ICP_QUERY: [define your target criteria]
- SCORE_THRESHOLD: 60
- CRM: hubspot

## Steps
1. Scrape companies matching ICP_QUERY from TheirStack. Save to leads-raw.csv.
2. Enrich leads-raw.csv via SyncGTM waterfall. Save to leads-enriched.csv.
3. Score leads-enriched.csv using scoring rules. Save to leads-scored.csv.
4. Push contacts scoring >= SCORE_THRESHOLD to CRM. Log results to leads-log.txt.
5. Report: total scraped, enriched, scored, pushed, skipped.

Run it with claude code run lead-agent.md. The agent executes all five steps sequentially, handling errors at each stage without halting the whole pipeline.

Schedule it via cron for daily runs:

0 7 * * 1-5 claude code run /path/to/lead-agent.md >> /logs/lead-agent.log 2>&1

Your team wakes up to a fresh batch of scored, enriched, CRM-ready contacts every weekday morning — with no manual work involved.

Pair this pipeline with Claude Code sales cadence automation to trigger personalized email sequences the moment each lead lands in your CRM.

Costs and Limits

At 500 leads/week, costs break down like this:

The main constraints are enrichment coverage (not every contact resolves to a verified email) and LinkedIn's scraping rate limits. For LinkedIn-heavy ICPs, rotate browser profiles and use a Playwright wrapper with session management.

Context window size is the other limit — CSVs over 1,000 rows should be chunked into batches of 200 before scoring or routing. Specify chunking in your prompt and Claude Code handles it automatically.

For a detailed breakdown of what to automate and what to keep human, read Claude Code sales automation: what to automate and what to keep human.