How to Build a LinkedIn AI SDR Agent with Claude and Item 12 MCP
· how-to
Technical SDRs and RevOps teams can skip the visual builder and let Claude design the agent. Item 12 MCP exposes workflow_create, workflow_list_blocks, and contact_enroll_workflow — Claude reads your ICP description, assembles steps, deploys the agent, and enrolls test contacts. Every send still queues for human review.
Prerequisites
- Item 12 account with LinkedIn connected in Settings
- MCP token created in Settings → ChatGPT / Claude / MCP
- Claude Desktop connector: URL https://item12.com/mcp, Bearer auth
- Read How to Connect Item 12 MCP to Claude Desktop if not set up yet
Step 1: Describe your ICP and sequence to Claude
Example prompt: "Build me a LinkedIn SDR agent for VP Sales at Series B SaaS in the US. Sequence: connection request with a one-line hook, wait 3 days after accept, then a 2-sentence DM asking about pipeline visibility. Use AI prompt mode on the DM step. Tag enrollments outbound-prospect-q3."
Step 2: Claude calls workflow_list_blocks → workflow_create
Claude first calls workflow_list_blocks to see available step types (connection request, wait, send message, conditions). Then workflow_create assembles the agent with step config — including ai_prompt on the DM block and merge token references. The returned workflow ID is your deployed agent.
Step 3: Test with contact_enroll_workflow
- Pick one test contact by profile URL or contact ID
- Ask Claude: "Enroll [profile URL] in workflow [name]"
- Claude calls contact_enroll_workflow
- Open Item 12 Review queue — approve the pending connection note or DM draft
Step 4: Iterate — adjust prompts and add follow-up steps
Refine via conversation: "Add a 7-day wait and a follow-up bump if no reply" or "Tighten the AI prompt to under 300 characters." Claude updates the workflow config. Re-enroll a fresh test contact to validate the new step order before bulk enrollment.
Step 5: Morning ops — review queue via MCP
- "List pending review items in my workspace"
- "Show me AI-drafted DMs waiting for approval from the Enterprise SDR agent"
- Approve in the Item 12 web app — MCP queues writes; approval UX lives in Review
MCP vs visual builder
Use MCP when you want Claude to draft sequence logic from natural language or iterate quickly across many variants. Use the visual builder when you prefer drag-and-drop and step-by-step UI. Both produce identical agents and share the same review queue.
Next in series
Close the loop with automated prospect discovery. Continue with How to Build an AI SDR That Finds and Messages LinkedIn Prospects Automatically.
Does Claude auto-send LinkedIn messages via MCP?
No — contact_enroll_workflow generates queued actions. You approve in Review before anything sends.
Do MCP workflow tools use credits?
MCP tool calls consume workspace credits per Item 12 plan. workflow_create and enroll are lightweight; search and enrich cost more.
Can ChatGPT or Cursor build agents the same way?
Yes — same MCP tools at item12.com/mcp with Bearer auth. This tutorial uses Claude Desktop; connector setup differs slightly per client.