WarmySender MCP Server — Connect Your AI to Cold Email, LinkedIn Outreach, and Warmup

Can Claude send cold emails for me?

Yes — Claude (or ChatGPT, Cursor, Codex, OpenClaw, Hermes Agent) can build, launch and manage your cold email, LinkedIn, Instagram and WhatsApp campaigns in WarmySender. It never sends a message itself: it writes the campaign and hands it to WarmySender's scheduler, which paces every send inside safe limits, and it can never raise a limit.

What this is

WarmySender is a B2B outreach platform that runs on autopilot with AI agents, across every channel: cold email, email warmup, LinkedIn, WhatsApp, Instagram, and multichannel sequences, plus real-time email verification. Our public MCP server lets any Model Context Protocol-compatible AI client run your outreach with natural language.

Point a supported client (Claude, ChatGPT, Cursor, Codex, OpenClaw, and Hermes Agent — plus Claude Desktop, Claude Code, Grok Bot, Windsurf, Zed, and anything else that speaks MCP 2025-11 Streamable HTTP) at https://warmysender.com/mcp with a Bearer API key, and your assistant gets 78 tools covering your whole workspace: create, launch, and manage campaigns on every channel — cold email, LinkedIn, and Instagram — publish and manage LinkedIn posts, run LinkedIn recruiter jobs (create, publish, close, and read applicants), enroll prospects, verify email addresses (single or batch), tune warmup, check mailbox and account health, and update suppression lists. Creating or launching a campaign only writes it and hands it to the same safety-checked scheduler our web UI uses; nothing sends in a burst.

Why connect your AI assistant?

Ask your assistant questions like:

• “List every email campaign that's running and show me reply rates for the last 7 days.”
• “Which of my mailboxes have a warmup score below 70?”
• “Enroll these 30 prospects into the Q2 Partnerships campaign.”
• “Pause the LinkedIn campaign for the Acme audience — the reply rate dropped this week.”
• “Publish a LinkedIn post announcing our new launch, and show me the engagement in a couple of days.”
• “Create and publish a LinkedIn job posting for a Senior Account Executive, then list the applicants as they come in.”
• “Pause the Acme Instagram DM campaign and tell me how many replies it got.”
• “What's my remaining LinkedIn invite budget on each account today?”
• “Add [email protected] and all of acme.com to my suppression list.”

Every action runs through the same scheduler-first path as the web UI. Rate limits, warmup ramps, LinkedIn daily/weekly caps, and domain-level protections all apply identically. MCP is a protocol adapter, not a parallel implementation.

Quick start in 3 steps

1. Create an API key. In WarmySender, go to Settings → API Keys, click Create, and pick the scopes you want to expose. Available scopes are campaigns:read, campaigns:write, prospects:read, prospects:write, mailboxes:read, warmup:read, warmup:write, linkedin:read, linkedin:write, enrollments:write, suppressions:read, suppressions:write, and jobs:read. Save the key value — you'll only see it once.
2. Paste the snippet into your client. Pick your AI client below, copy the install snippet, and replace ws_YOUR_API_KEY_HERE with your real key.
3. Ask your assistant to use WarmySender. Try “List my campaigns” or “What's my LinkedIn account health?” — the client discovers MCP tools automatically.

Install snippets

Server URL: https://warmysender.com/mcp. Transport: Streamable HTTP (MCP spec 2025-11-25). Replace ws_YOUR_API_KEY_HERE with your API key.

{
  "mcpServers": {
    "warmysender": {
      "command": "npx",
      "args": [
        "-y",
        "@warmysender/mcp",
        "--api-key=ws_YOUR_API_KEY_HERE"
      ]
    }
  }
}

claude mcp add warmysender -- npx -y @warmysender/mcp --api-key=ws_YOUR_API_KEY_HERE

{
  "mcpServers": {
    "warmysender": {
      "url": "https://warmysender.com/mcp",
      "headers": {
        "Authorization": "Bearer ws_YOUR_API_KEY_HERE"
      }
    }
  }
}

{
  "mcpServers": {
    "warmysender": {
      "serverUrl": "https://warmysender.com/mcp",
      "headers": {
        "Authorization": "Bearer ws_YOUR_API_KEY_HERE"
      }
    }
  }
}

{
  "context_servers": {
    "warmysender": {
      "command": {
        "path": "npx",
        "args": [
          "-y",
          "@warmysender/mcp",
          "--api-key=ws_YOUR_API_KEY_HERE"
        ]
      }
    }
  }
}

Available tools

MCP tools are gated by the scopes on your API key. If a call returns insufficient_scope, add the missing scope in Settings → API Keys and create a new key. Every write tool accepts an optional idempotency_key argument so retries are safe. Destructive tools (pause_*, unenroll_*, add_to_suppression_list) set destructiveHint: true so MCP clients prompt the user for confirmation before calling.

Safety & account protection

WarmySender's MCP server is designed so that a compromised or confused AI cannot burn your sending reputation or ban your LinkedIn accounts. Five guardrails always apply:

1. Workspace scoping on every call. Your API key resolves to exactly one workspace. Every query is filtered by that workspace id — there is no path to read or mutate another tenant's data.
2. Scheduler-first execution. LinkedIn invite/message/view calls never happen inline from MCP. The scheduler owns the rate limits, ramp schedule, and per-account safety caps. Enrolling 1,000 prospects at once still sends at the safe rate.
3. LinkedIn daily/weekly limits & 4-week ramp are non-negotiable. A new account starts at ~10 invites/day and grows over 4 weeks to safe maximums (50 invites/day, 150 messages/day). MCP cannot raise these — account safety always wins over speed.
4. Per-mailbox sending pacer and warmup [ref: XXXX] tag stay intact. MCP does not expose any tool that modifies or removes the warmup subject-line identifier or bypasses per-mailbox daily caps.
5. No account connect / disconnect / delete tools. Connecting and disconnecting email or LinkedIn accounts stays in the UI. No delete_campaign, delete_mailbox, or delete_account tools exist in V1 — soft-cancel via pause/archive only.

Read more about the LinkedIn safety model in our LinkedIn Safety FAQ.

Rate limits

Limits are per workspace, in a rolling 60-second window, and separate from your monthly call allowance. Every budget scales with your plan — read down your plan's column. These are in addition to — not replacing — the scheduler's hard LinkedIn, Instagram, and warmup caps.

Figures are calls per minute, per workspace. Agent actions start on the Pro plan — a Free plan has no agent allowance yet, so there is no per-minute budget to spend on one and calls come back pointing at where to change plan instead of running. Customers on the legacy Starter and Business plans keep their own pace; the AI Agents page lists every plan side by side.

Exceeding a per-minute budget returns a rate_limited error carrying retry_after_seconds, so your assistant knows when to retry. A plan refusal is a different answer — plan_upgrade_required, with no wait hint, because waiting cannot cure it.

Error codes

MCP tools use JSON-RPC error codes. The WarmySender server reserves the -32000…-32099 range for protocol-specific errors on top of the standard JSON-RPC codes.

Frequently asked questions

Related reading

• Full WarmySender documentation — setup guides, troubleshooting, and help for every feature
• Pricing — plans, LinkedIn seats, and annual billing
• REST API reference — all API v1 endpoints, scopes, and webhook events
• Settings & API Keys — create, scope, and revoke API keys
• Integrations hub — HubSpot, Salesforce, Zapier, Make, Slack, and more
• LinkedIn Safety FAQ — how we keep accounts safe
• Email Warmup FAQ — what warmup does and why you need it
• Model Context Protocol (modelcontextprotocol.io) — the open spec behind MCP