Lead Generation With Claude: Complete MCP Setup Guide (2026)

Lead generation with Clauderequires more than a language model — it requires a live data connection. Ask Claude to "find me 25 VP Sales contacts at Series B SaaS companies in the US" without one and it writes back plausible-sounding names that do not exist. The model generates text from patterns — it has no live database to query.

Connect the SyncGTM MCP and the same prompt runs a real search. Claude calls find_companies, pipes the domains into find_people, pulls work emails with find_work_email, verifies them with verify_email, and writes the output to a CSV or pushes it to HubSpot — all inside one conversation.

This guide covers the complete setup and end-to-end workflow: connector configuration in claude.ai and Claude Code, the exact prompts to use at each step, sample output for every tool call, and the credit cost so you can estimate spend before running anything.

Why Use Claude for Lead Generation?

Most AI lead generation tools wrap a database in a chat interface and call it an agent. What makes Claude different is the Model Context Protocol — an open standard Anthropic published so any external service can expose structured tools that Claude calls natively, with the same reasoning it applies to everything else.

In practice, that means three things you cannot get from a dedicated prospecting UI:

• Chaining without exporting. The output of one tool becomes the input to the next inside the same conversation. Company domains from find_companies feed straight into find_people — no CSV in between.
• Natural-language filter refinement."Too many staffing companies — drop them and add a minimum of 50 employees" is one sentence, not a rebuilt filter set and a second export.
• Output in any format. The same search can produce a markdown table, a CSV file, a CRM payload, or a written account brief. The data never has to land in a spreadsheet first.

The catch is that all of this only works when Claude is actually calling a tool. Without a connected MCP, it answers from training data, and training data does not have your target accounts' current headcount, last funding round, or verified email addresses.

The SyncGTM MCP Tool Set

The SyncGTM MCP server (documented in the SyncGTM MCP docs) exposes more than 40 tools, but four handle most of a standard lead generation run. Here is how they map to workflow stages and what each costs.

Search tools are billed per run, not per row. A find_companies run returning 25 results costs the same 3 credits as one returning 100. The credit lever that matters is on the enrichment side: every find_work_email and verify_email call is billed per person, so narrow your list before you enrich it.

The full server also covers company enrichment, email finding across multiple providers, LinkedIn post engagement, job opening signals, and tech-stack detection — but those are add-ons once the core workflow runs cleanly.

Step 1 — Connect the SyncGTM MCP to Claude

The SyncGTM MCP is a remote HTTP server authenticated via OAuth. No API key to paste, no .env file to manage, no token to rotate. Setup differs slightly between the claude.ai web interface and Claude Code.

Option A — claude.ai (Connectors)

This path works inside any claude.ai conversation, including Claude on mobile. The connector persists across your account — you set it up once.

1. Sign in to claude.ai in your browser.
2. Go to Settings → Connectors → Add custom connector.
3. Paste the connector URL: https://api.syncgtm.com/mcp
4. Click Save. Claude opens a browser tab to the SyncGTM OAuth sign-in.
5. Sign in to your SyncGTM account. On the approve screen, confirm you are authorizing the correct organization — if your account belongs to multiple workspaces, make sure the right one is selected before clicking Approve.
6. Return to claude.ai. The connector should now show as connected. Start a new conversation to activate it.

Option B — Claude Code (terminal)

Use this path if you run Claude Code in the terminal and want the MCP available in agentic scripts as well as chat. See the B2B database MCP setup guide for the full Claude Code walkthrough.

# Add to the current project (scoped to this repo):
claude mcp add --transport http syncgtm https://api.syncgtm.com/mcp

# Or add globally across all projects:
claude mcp add --scope user --transport http syncgtm https://api.syncgtm.com/mcp

# Restart Claude Code, then authenticate:
/mcp
# Select: syncgtm -> Authenticate
# Browser opens SyncGTM OAuth. Sign in and approve.

After authenticating, run /mcp again. The syncgtm entry should show as connected with its tools listed. If it shows disconnected, re-run the add command and make sure you are signed in to SyncGTM in your default browser — the OAuth handoff targets that one.

Step 2 — Check Your Credits Before Anything Else

check_credits is free and takes one second. It also confirms the MCP connection is live — if Claude answers with a number, the tools are working. If it says it cannot access SyncGTM, fix the connection before continuing.

# Exact prompt to use:
"Check my SyncGTM credit balance."

# Claude calls check_credits and responds with something like:
# Your SyncGTM balance is 142 credits.

Before any large run, estimate your spend: a 25-lead end-to-end workflow costs roughly 38–39 credits (3 + 3 + 25 + 7.5). If your balance is low, top it up on the SyncGTM pricing page before you start, or cap the list at fewer contacts.

Step 3 — Define Your ICP for Claude

Claude maps a natural-language description onto the filter parameters of find_companies. The more specific you are, the tighter the list — and the less enrichment spend you waste on companies that do not fit.

A good ICP prompt covers at least six dimensions:

The exclusion side is equally important. Give Claude a list of domains for your existing customers and the industries you never close. Every row excluded at this stage saves credits downstream on enrichment.

Step 4 — Find Target Accounts With find_companies

find_companies is the account-discovery tool. It costs 3 credits per run and returns up to 100 company profiles per call. You describe the accounts in plain English and Claude maps the description onto the available filters.

Exact prompt

"Use find_companies to find 25 B2B SaaS companies in the US or Canada
with 50–500 employees that raised a Series A or Series B in the last
24 months and run HubSpot or Salesforce. Exclude staffing, recruiting,
and consulting. Return a table with: domain, company name, headcount,
last funding round, HQ country."

What the tool call looks like

In the claude.ai UI you will see a tool-use block before the result — something like find_companies called with a JSON payload showing the mapped parameters. This is the confirmation that real data was fetched. No tool-use block means Claude answered from memory — see the memory trap section.

Sample output

Credit cost:3 credits, regardless of whether you get 5 results or 25. If the initial list is too broad, refine in the same conversation — "Drop anything under 100 employees and add a minimum of $5M in total funding." Refinement is free — it uses the context Claude already holds.

Step 5 — Find Decision-Makers With find_people

find_people searches profiles by title, seniority, function, geography, and timing signals like recent job changes. You can pass the domains from the previous step directly — Claude carries them in context without you copying anything.

Exact prompt

"Use find_people to find VP-level or above contacts in Sales or
Revenue Operations at the 25 companies from the last search.
Based in the US. Limit to 25 results. Return: name, title,
company, domain, LinkedIn URL."

Claude passes the domains into current_company_domains and combines seniority_levels (VP and above) with current_functions (Sales, Revenue Operations) and person_countries (US).

Sample output

Credit cost: 3 credits for the run. Two useful modifiers to add: pass recently_changed_jobs: true to surface new buyers with fresh budget authority, or has_verified_business_email: true to pre-filter to people whose email is already verified — which cuts your find_work_email spend in the next step. Just add the modifier to the prompt in plain English and Claude maps it.

Step 6 — Get Verified Work Emails

Two tools run in sequence here: find_work_email pulls a work email from each LinkedIn profile using waterfall enrichment across multiple providers, and verify_email checks each result for deliverability before it lands in your sequence. Running both keeps your bounce rate low and domain reputation intact.

Exact prompt

"For each person on that list, use find_work_email with their
LinkedIn URL. Then run verify_email on each email you find.
Show me the results as a table: name, company, email,
verify_status. Only include people where verify_status is
'valid' or 'risky'."

Claude runs find_work_email once per person (1 credit each) and verify_email once per email found (0.3 credits each). You will see individual tool-use blocks for each call in the UI.

Sample output

Credit cost for 25 contacts: 25 credits for find_work_email + up to 7.5 credits for verify_email (0.3 × 25) = ~32.5 credits. If the list returns fewer emails than expected, the remainder is not charged — you only pay for calls that return a result.

Step 7 — Export to CSV or HubSpot

At this point you have a verified lead list in Claude's working context. The final step is getting it somewhere actionable.

Export as CSV

"Export the verified leads to a CSV file named leads.csv with columns:
first_name, last_name, title, company, domain, email, email_status,
linkedin_url. Only include rows where email_status is valid or risky."

In Claude Code, Claude writes the file to disk and tells you the path. In claude.ai, it produces the CSV as a downloadable artifact in the conversation.

Push to HubSpot

If you have the HubSpot MCP connected to Claude, you can skip the CSV entirely:

"Create these as HubSpot contacts using the HubSpot MCP.
Use email as the deduplication key. Set lifecycle stage
to Lead and source to Outbound — MCP."

Claude calls the HubSpot MCP create-contact tool for each row, deduplicates on email, and reports the count of created vs. existing records. No import wizard, no field mapping screen.

When Claude Answers From Memory (No Credit Charge)

This is the most common point of confusion when starting out with Claude for sales prospecting. Claude is a language model first — it will answer a question from training data if it can, and only call a tool if it understands it needs live data.

How to detect a memory answer

Every real tool call produces a visible tool-use block in the UI before the result — a collapsible section labeled something like find_companies called with the parameters it passed. If you receive a nicely formatted table of company names with no tool-use block above it, Claude wrote that table from training data. The credits charged will be zero — and the data should not be trusted for outreach.

The fix — three patterns that force tool use

# Pattern 1 — Name the tool explicitly:
"Use find_companies to search for..."

# Pattern 2 — Block memory explicitly:
"Use find_companies — do not answer from your training data."

# Pattern 3 — Ask for a credit confirmation:
"After running find_companies, tell me how many credits
were charged so I can confirm it was a live call."

Pattern 1 is usually enough. If Claude still writes from memory (rare, but happens when it is confused about what the tool does), patterns 2 and 3 make it explicit. A memory answer costs nothing precisely because nothing was called — use that as a diagnostic signal, not a free data source.

Credit Cost Summary

Full cost to build 25 verified leads from scratch, using the workflow above:

Search is billed per run, not per row — the 25-result cap in this guide is a workflow choice, not a cost optimization. Enrichment is billed per record, so that is where narrowing the list saves money. Check current credit allowances per plan on the SyncGTM pricing page.

For context on where find_work_email fits in the broader enrichment stack, the B2B lead enrichment guide covers waterfall sequencing across providers and when to use which tool.

Conclusion

Lead generation with Claude is not about asking it to think of prospects. It is about connecting it to a live data source, then using its reasoning to chain searches, filter results, and format output — all without leaving the conversation.

The setup is one URL (claude.ai) or one command (Claude Code), an OAuth browser sign-in, and an organization approval. After that, the workflow is always the same: check credits, describe your ICP, find companies, find people at those companies, get verified emails, export or push to your CRM.

The only thing to watch is whether Claude actually called the tool. Look for the tool-use block. If it is there, the data is real. If it is not, add the phrase "use find_companies" to your prompt and run it again. Create a free SyncGTM account to connect the MCP and run your first search today.