Lead Friendly for AI agents

Lead Friendly runs a public MCP server (Model Context Protocol). Point any MCP-capable agent — ChatGPT, Claude, or your own orchestrator — at the endpoint below and it can ask what Lead Friendly costs, which industries it serves, whether it can do a given job, and hand off a lead or a demo request. Agent to agent, no human in the loop.

Customers get a second, authenticated server that acts inside their own account —see below, or the connector guide if you want it wired up rather than built against.

For the overview — every tool on both servers, the safety model and what it costs — see the MCP server page.

Endpoint

POST https://www.leadfriendly.com/api/mcp
Content-Type: application/json
Transport: Streamable HTTP (JSON-RPC 2.0)
Auth: none (public discovery + lead-gen)

Discovery descriptors: /.well-known/mcp.json · /.well-known/ai-plugin.json · /llms.txt

Tools

get_pricing
Current subscription plans and prices, live from the source of truth.
list_industries
The industry verticals with a tailored AI agent template (optional keyword filter).
check_capability
Whether Lead Friendly can do a specific job — e.g. "call leads after hours", "dispatch trucks".
book_demo
Request a product demo for a prospect; returns a self-serve booking link.
create_lead
Hand off an interested sales lead to the Lead Friendly team. Never touches a customer’s CRM.

Try it

List the available tools:

curl -s https://www.leadfriendly.com/api/mcp \
  -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Call a tool (get pricing):

curl -s https://www.leadfriendly.com/api/mcp \
  -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/call",
       "params":{"name":"get_pricing","arguments":{}}}'

The workspace server (authenticated)

The server above is public and sells. The one below is the product. A Lead Friendly customer can hand their own AI assistant a key and let it work inside their dealership account — read the CRM, text a customer, have an AI voice agent place a call, book the appointment — through exactly the same compliance and billing machinery a salesperson goes through.

POST https://www.leadfriendly.com/api/mcp/v1
Authorization: Bearer lf_live_…
   (or X-LF-Api-Key: lf_live_… if your client reserves Authorization)
Transport: Streamable HTTP (JSON-RPC 2.0)
Protocol: 2025-06-18

Descriptor: /.well-known/mcp-v1.json

Getting a key

In Lead Friendly, go to Settings → API keys and create one. Pick “Read only” to let an agent look without touching, “Read + setup” to let it build agents, campaigns and automations (all switched off) without contacting anyone, or “Read + write” to let it text, call, book and update as well. The key selects the account, so every user connects with their own and an agent can never see another dealership's data. Revoking is one click and takes effect on the next request.

Connecting

In Claude Code:

claude mcp add --transport http lead-friendly   https://www.leadfriendly.com/api/mcp/v1   --header "Authorization: Bearer lf_live_…"

Or straight over HTTP:

curl -s https://www.leadfriendly.com/api/mcp/v1   -H 'Authorization: Bearer lf_live_…'   -H 'Content-Type: application/json'   -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Setting this up for a store rather than for yourself? The step-by-step connector guide walks a non-developer through the same tools in the Claude, ChatGPT and Gemini interfaces — which plans support it, who sets it up, and which tools to leave on “needs approval”.

Things to ask it

“Is my account clear to text customers right now? If not, what is missing?” — get_org_status
“Find everyone who asked about a truck in the last month, tell me which of them I'm allowed to text, and text those.” — search_contacts, get_contact, send_sms
“What is still unsigned on the Alvarez deal, and when are they coming in?” — list_deals, get_deal, list_appointments
“Call my two hottest leads with Maya and book whoever answers.” — start_outbound_call, book_appointment

What it will refuse

An agent gets the account's guard chain, not a way around it: 10DLC registration, quiet hours in the customer's time zone, do-not-call, STOP opt-outs, AI disclosure, recording disclosure, and the wallet. On top of that there is one rule that exists only for agents — marketing outreach requires a consent record on the contact, and a claim that a message is transactional is checked before it is believed: against an appointment the agent did not book itself, a deal still in progress, or a recent inbound message, and against the wording, so a message that reads as an offer is refused however it is labelled. An agent cannot override a compliance warning; that takes a signed-in human.

Refusals come back as ordinary tool results — ok:false with a code, a reason and a remediation — so the assistant explains what to fix instead of retrying. Every call, including every refusal, is written to the account's agent activity log under Settings → AI agents. Arguments are hashed, never stored: the message body and the customer's number are not copied into an audit table.

Safety: on the public server every tool is read-only or routes an inbound lead to the Lead Friendly sales team — nothing there authenticates a user, reads tenant data, or writes into any organization's CRM, and it is rate-limited per IP. The workspace server reaches customer data only for the one account its key belongs to, and never returns social security numbers, dates of birth, driver's licences, income or credit scores to any agent. See our privacy policy.

Questions?

Email hello@leadfriendly.com or read the compliance posture before wiring outbound.