The AI-driven SDR process: MCP flow
The platform has two modes. This guide covers the first one: you bring your lists, LDM handles delivery, and the whole process is driven by your AI agent or a chat with an MCP connector. From sign-up to replies in the Inbox.
Work the way you like
There was a wave when everyone lived in chats. Now people are moving to agents. For us it's the same thing: the MCP connector plugs into a chat (Claude Desktop, Perplexity) and into an autonomous agent (Claude Code, Cursor, any MCP client) alike. A person in a chat gives tasks in plain words; an agent does the same on its own.
Either way, the same LDM sits on the other side: the same tools, the same expert layer, the same legal guardrails. The interface is a matter of habit, not capability.
Chat or agent doesn't matter. What matters is that the process follows a proven flow, not gut feeling.
The LDM MCP agent doesn't just execute — it accompanies
However smart your agent is, it doesn't see the whole context: deliverability state, domain history, the typical pitfalls of each stage. So LDM answers not as a bare API but as a market expert: every response carries an _expert block — what to do next, the typical flow, where people usually fail.
Think of it as a consultation: a lawyer and a doctor can each be brilliant alone, but exchanging opinions in the moment prevents expensive mistakes. Your agent makes the decisions — our layer keeps reminding it what's going on around. The capability map and end-to-end scenario live at /api/v1/agent-guide; the agent reads it itself.
A person can do all of this alone. Our job is to keep them — and their agent — from failing where everyone fails.
From zero to replies — eight steps
An example workflow: begin with 5–10 companies and one message, then approve the next step. This is a walkthrough, not a promise of leads.
Every stage has a legal layer. You don't have to remember it: the platform won't let you skip it.
You can start right away
- 01Your offer, ideal customer profile, geography, exclusions, supporting facts and any presentation. Agree what counts as a useful result before selecting companies.
- 02Registration starts with PENDING. Reads are available; sending and other permissions require activation. Depending on your email and website, activation may require a questionnaire and review.
- 03Once the workspace and creative permissions are ready, LDM can send the test copy and linked attachment to the workspace owner’s registration email. The agent uses that address automatically, without asking for it again.
- 04This sends materials to you. Campaigns to prospects require a connected sending account or pool and a separate readiness check. A sample sent by LDM does not test your own sending domain.
- 05Delivery billing: sent + 24 hours without a recorded bounce. This is not proof of the recipient’s Inbox folder. One prepaid balance for web UI, API and MCP. Delivery and preparation are separate operations. The table loads the current public USD rates; your workspace may have individual rates.
Agree the first result and permissions before execution. Activation and review can take more than one session.
Connect right now
Copy the config for your client — or just hand the agent a ready-made prompt and it does the rest.
- Claude Desktop — claude_desktop_config.json — the snippet below. Restart the app and the LDM tools appear on their own.
- Claude Code — One command: claude mcp add ldm -e LDM_API_KEY=ldm_… -- npx -y --package=@live-direct-marketing/sdk@latest ldm-mcp
- Perplexity (Mac) — Settings → Connectors → Add Connector → Simple tab: the same npx command with the key in env.
- Cursor and other MCP clients — The same snippet in the client's MCP servers config.
{
"mcpServers": {
"ldm": {
"command": "npx",
"args": ["-y", "--package=@live-direct-marketing/sdk@latest", "ldm-mcp"],
"env": { "LDM_API_KEY": "ldm_..." }
}
}
}Prompt for your agent — copy and paste
Connect to LDM over MCP and read the agent guide. Check my workspace status and permissions. Ask for my offer, ideal customer profile, geography, exclusions and supporting materials. Prepare 5–10 companies with sources and reasons, then one sample message for my review. If I have no sending account, use the platform test-send to my registration email without asking for my address again. Show costs and readiness. Connect a campaign sender separately; launch only after my confirmation.
https://api.live-direct-marketing.online/mcpThese clients can’t spawn a local process — connect them to our hosted server at https://api.live-direct-marketing.online/mcp. In claude.ai / Claude Desktop add it as a custom connector with a request header Authorization: Bearer ldm_.... In ChatGPT (developer mode), where custom headers aren’t supported, use the personal URL https://api.live-direct-marketing.online/mcp/k/<your-key> with authentication set to None — and treat that URL as a secret.
Documentation
- Developer portal — developers.live-direct-marketing.online — API, keys, MCP, A2A.
- Platform help — live-direct-marketing.online/help — the whole interface, section by section.
- Agent map — /api/v1/agent-guide — the end-to-end scenario, read by agents automatically.
A legal layer at every step
The “your lists — our delivery” mode works white-hat only. The rules are built into the conveyor:
- 01Lists only with a lawful basis: your own leads, public corporate contacts, confirmed interest. Source confirmation is a mandatory step.
- 02An honest sender, a working unsubscribe link in every email, opted-out addresses in suppression forever.
- 03Stop lists and suppression are checked automatically before every send.
- 04Personal data handled by the law (GDPR and local equivalents): minimal data, deletion on request.
We won't send what must not be sent — even if the agent asks.