Automate your funnel creation and analysis with your own AI agents.
The LeadCapture MCP server gives Claude, Codex, Cursor, and any MCP-compatible agent a direct, permissioned connection to your account. Describe what you want in plain language. Your agent builds funnels, restyles pages, remaps lead delivery, runs A/B tests, and audits your whole operation.
“Claude has totally redesigned my funnel! Amazing, total game changer. If it ever gets stuck it tells me exactly what to do next. It's unreal!”
What is the LeadCapture MCP server?
LeadCapture.io has an MCP server. It connects Claude, OpenAI Codex, Cursor, and any MCP-compatible agent directly to your LeadCapture account through 49 tools covering funnels, steps, fields, styling, conditional routing, lead scoring, lead delivery, integrations, analytics, A/B experiments, and lead records. You connect it once with a scoped API key. After that, you describe what you want in plain language and your agent does the work in your account: new funnels, new steps, routing logic, delivery fixes, experiments, and audits. It is included on every plan.
And you get the best of both worlds. Copilot is the AI assistant built into the LeadCapture dashboard. The MCP server is the door for your own agents: Claude in your terminal, Codex in your editor, a workflow you wrote yourself. Build with our assistant in the app, bring your own, or both. Either way, you can build agents on top of your lead operation.
Copilot is your in-app expert. MCP is your automation layer.
Copilot
Copilot lives inside the builder. It knows the funnel you are looking at, answers questions, and builds alongside you while you work. When you are in the app, it is the fastest way to get something done.
MCP server
The MCP server is for your own agents: Claude in your terminal, Codex in your editor, Cursor in your IDE, or a workflow you wrote yourself. They build, edit, and analyze your funnels from wherever they run, whether or not you are logged in.
Both run the same engine. A funnel your agent builds arrives fully built: steps, fields, routing, styling, and delivery. Not a scaffold you finish by hand.
What can an agent do with the LeadCapture MCP server?
Forty-nine tools, grouped by the job you hand it.
Build funnels
Create a funnel from a description, add and reorder steps, add and edit fields, clone what already works.
- create_funnel
- add_step
- add_field
- update_step
- update_field
- reorder_steps
- duplicate_step
- clone_funnel
Design and style
Brand colors, button and field styling, teaser text and reassurance blocks, landing page elements.
- update_funnel_styles
- set_step_teaser
- update_landing_element
- reorder_landing_elements
Logic and qualification
Conditional routing, lead scoring, in-form calculators and computed values.
- set_step_routing
- set_step_calculations
- upsert_calculation_variable
- update_funnel_settings
Lead delivery
Inspect where leads go, remap fields, send test payloads, and check whether delivery is failing.
- list_integrations
- update_field_mapping
- test_integration
- get_integration_health
Measure and optimize
Funnel and account analytics, drop-off, and the full A/B testing lifecycle.
- get_funnel_analytics
- get_account_analytics
- audit_account
- start_experiment
- get_experiment_results
- end_experiment
Leads and support
Read full lead records, verification status and Contact Score included, and answer product questions from our help docs.
- list_leads
- get_lead
- search_help_docs
Prompts that work
Every prompt below works as written. Connect your account, paste one into Claude, Codex, or Cursor, and watch it happen.
Fix and refine
the standing instruction> Make every zip code field US-only with strict 5-digit validation, across all my funnels.> The phone step is losing people. Show me the drop-off, then rewrite that question and add a reassurance line under the button.> Restyle this funnel to match my client's brand: navy primary, white buttons, rounded corners.Launch something new
the demo moment> Build me a solar quote funnel: postcode, roof type, monthly bill, then name, email and phone. Route anyone renting to a disqualified page.> Build a lead quiz that scores each homeowner on solar readiness and routes anyone below 60 to a nurture page.> Clone my best-performing funnel and rewrite the copy for the Texas market.Understand performance
the question you cannot click> Which of my funnels converted worst last month, and what do the drop-offs have in common?> Audit my account and tell me what to fix first.Keep delivery healthy
the money path> Is anything failing to deliver right now?> My buyer renamed their zip field to postal_code. Remap it and send a test lead.Run experiments
for the growth crowd> Set up an A/B test on my headline, split it 70/30, and tell me when it has a winner.For agencies
every client, one session> Copy my winning roofing funnel into every client workspace that runs home services offers.> Pull last month's conversion rate for every client account and rank them.Turn prompts into standing workflows
Pair the MCP server with Claude Code and the routine stops being yours to remember. Claude Code chains LeadCapture tools with everything else it can reach: your buyer's field specs, your ad account exports, your own scripts. Your agents stop being something you talk to and become something that maintains your lead operation.
Audit my account and flag anything high severity.Pull last week's funnel analytics and write me the summary.Watch integration health, and when my buyer changes their spec, remap the fields and send a test lead.Your whole stack, one analysis
Your agent is not limited to LeadCapture. Connect it to Meta, Google Ads, and GA4 alongside the MCP server and it can join ad spend with funnel performance: cost per qualified lead by campaign, drop-off against creative, one report across everything you run. Your data, your agent, your dashboard.
> Pull yesterday's spend from Meta and Google Ads, join it with each funnel's qualified lead count, and tell me my cost per qualified lead by campaign.Is it safe to let an agent touch live funnels?
Yes, and not because we ask you to trust it. Four protections, all enforced in code.
Nine separate permissions. Read-only is a real option, and lead data is its own scope you can withhold. An existing Data API key does not silently gain agent access.
Every agent change is snapshotted before it happens and restorable from the funnel's version history.
Anything a visitor would notice, including changing where leads are delivered, requires your explicit confirmation.
Your agent can read, test and remap an integration without ever seeing its API keys or tokens.
API key scopes · the nine permissions
Frequently asked questions
Does LeadCapture.io have an MCP server?+
Yes. LeadCapture.io has an MCP server at my.leadcapture.io/api/mcp. It gives Claude, Codex, Cursor, and any MCP-compatible agent permissioned access to your account through 49 tools, and it is included on every plan.
Can I build lead funnels with Claude?+
Yes. Describe the funnel in plain language and Claude builds it in your LeadCapture account: steps, fields, conditional routing, lead scoring, styling, and delivery. The same works in Codex, Cursor, or any MCP client.
Can AI agents create lead quizzes and multi-step forms?+
Yes. Agents build multi-step forms, lead quizzes with scoring and calculators, and landing pages, and they qualify what those capture. A qualified lead is one with a verified email, a verified phone, enriched data, and a Contact Score above your threshold, checked before the visitor leaves the page.
Which AI tools work with the LeadCapture MCP server?+
Claude, Claude Code, Claude Desktop, OpenAI Codex, Cursor, and any client that supports the Model Context Protocol. They all connect the same way: one endpoint, one scoped API key.
Is it safe to let an AI agent change live funnels?+
Yes. Permissions are scoped across nine separate grants, every change is snapshotted and restorable, anything visitor-facing requires your confirmation, and integration credentials are never exposed to the agent.
How is the MCP server different from Copilot?+
Copilot is the expert inside the LeadCapture builder. The MCP server connects your own external agents, so they can build, edit, and analyze funnels from your terminal, your editor, or your own workflows. Both run the same underlying engine.
Does it work for agencies managing client accounts?+
Yes. One connection reaches every client workspace, so an agency can roll out a winning funnel across clients, pull analytics for every account, and monitor delivery health from a single agent session.
What does the MCP server cost?+
Nothing extra. It is included on every LeadCapture plan, from Starter at $199/mo to Enterprise. No add-on tier and no AI credit meter. Plans come with a 7-day free trial, no credit card required.
What is an MCP server?+
MCP (Model Context Protocol) is the open standard AI agents use to operate software on your behalf. An MCP server is the connection point a product exposes. LeadCapture's MCP server is the connection point for your lead operation.
How do I connect Claude to LeadCapture?
Create an API key in Settings and choose its scopes. Read-only is a real option.
Add the endpoint my.leadcapture.io/api/mcp to Claude, Codex, Cursor, or any MCP client.
Ask for something. “Audit my account and tell me what to fix first” is a good place to start.
Same traffic. Capture more qualified leads.
34% average conversion lift across the customer base. Now your agent works the same numbers.