Build a lead-generation agent with MCP and ScanToPitch

An 'agent' is just an AI that can take several steps toward a goal on its own, deciding what to do next based on what it finds. With ScanToPitch's MCP tools, you can build one that runs a real lead-generation loop. Here is the blueprint.

Give it a goal, not a script

Instead of 'do step one, then step two,' you hand the agent an outcome: 'produce five well-qualified leads for web-redesign outreach in a city I name, each with a drafted first email.' The MCP tools let it choose the steps, scan, read results, filter, generate, to reach that outcome.

The tools it has

Through MCP the agent can start area scans, scan single URLs, check scan status, read scored leads, and generate mail and call scripts, all against your account. It can also list your past scans, so it can build on earlier work rather than starting cold every time.

The loop it runs

• Scan the market you named and wait for results.
• Read the leads and rank them by website weakness and reachability.
• Drop chains and sites with no contact path.
• For the survivors, pull the specific problems and draft a grounded first email.
• Return the shortlist and drafts for your review.

The guardrails that come for free

You do not have to build safety in; ScanToPitch enforces it. Every tool call runs through the same checks as the app: credit debits, rate limits, trial caps and account suspension all apply. Failed scans refund automatically. And purchases are human-only, so an agent can spend credits doing work but can never buy more. That means an experimental agent can misfire without costing you money.

Where to run it

For a human-supervised agent, an MCP client like Claude is enough. For a fully autonomous, scheduled agent, pair the same logic with the REST API. Either way the account, credits and limits are shared, so you can prototype in chat and productionise in code without rebuilding anything.