agent-cold-email

Purpose

Coldrig is a cold email infrastructure designed specifically for coding-oriented AI agents. Unlike humans clicking around on a SaaS dashboard, it lets agents (Claude Code, Codex, Cursor, Cline, or any tool that supports MCP and can call a CLI) drive the entire cold email process. After signing up, users only need to hand a bearer token to the agent, and it can automatically handle brand lookalike domain purchases, mailbox setup and warm-up, sequence delivery, reply handling, and metrics reporting.

Key Capabilities

• Provides 28 high-level intent tools (such as setup_infrastructure, launch_campaign, inbox, metrics, etc.), rather than simply exposing the vendor's 100+ tools.
• Supports both MCP and HTTP interfaces, and provides an npm CLI (agent-cold-email@0.2.2). The CLI includes an mcp subcommand that can act as a stdio bridge to the hosted /mcp endpoint, for MCP clients that only support stdio.
• Each customer has isolated domains and mailboxes, not shared with other tenants.
• Provides a free fault-injection sandbox demo (no signup, no real sending), which can run the full provision → warm → send → reply → report flow, and includes simulations of rate limits, bounces, timeouts, asynchronous replies, and more.
• Provides an optional, agent-configured dashboard and unified inbox (located at /app, where the agent can control its layout via MCP).
• Outbound webhooks (get_webhooks, configure_webhook): reply, bounce, soft bounce, complaint, and unsubscribe events are pushed with HMAC signatures to your own HTTPS endpoint.

Usage

• MCP (recommended for Claude Code / Codex): hosted MCP endpoint /mcp, JSON-RPC 2.0 over streamable HTTP.
• Codex CLI: configure ~/.codex/config.toml or the trusted project's .codex/config.toml, first setting COLDRIG_TOKEN. For Claude Code, Cursor, and Cline integration methods, see the official connection page.
• Plugins/Skills: The Claude Code plugin connects to the same MCP server and prompts for the bearer token; you can also install only the skill: npx skills add YS-projectcalc/agent-cold-email.
• CLI: npm package agent-cold-email@0.2.2, npx agent-cold-email demo can run the demo directly without a local build.
• HTTP: all 28 intents at https://api.coldrig.dev Provides a bearer token-protected REST interface that any HTTP client can drive directly; the complete REST contract is in the official openapi.yaml, and agent-facing documentation is in AGENTS.md.
• First use: POST /signup creates a demo plan tenant (the CLI demo command does this automatically, no manual registration needed); demo mode is structurally unable to reach real vendor adapters, and all operations are simulated in a fault-injection sandbox, costing nothing and never touching real inboxes.

Use cases

Suitable for tasks in which a coding agent evaluates, sets up, and runs cold email outreach campaigns, for example "set up and run a cold email outreach campaign." The platform handles infrastructure, sequences, and delivery guardrails; outreach content is written by the customer's agent itself, and the platform does not generate outreach copy or run an opaque "AI SDR."

Limits and compliance

• The per-mailbox daily sending cap (about 40–50 emails/mailbox/day) is enforced by the engine.
• Built-in suppression list and one-click unsubscribe (RFC 8058); unsubscribed addresses will not be contacted again; a complete CAN-SPAM opt-out process is provided.
• The real mailing address and sender identity collected from each tenant during setup are injected into every email footer.
• Automatic pause on complaint rate: mailboxes approaching Gmail's 0.30% unusable threshold are throttled or paused.
• Lookalike domains are limited to the sender's own brand; server-side code enforces this, hard-rejects a denylist of well-known brands, and requires the claimed brand to correspond to the primary domain. Not a phishing or identity impersonation tool.
• Warm-up is honestly described as multi-week reputation building, not "bypassing spam filters," and the platform provides no filter-evasion mechanism.
• There is currently no multi-year production or deliverability record; inbox placement or delivery performance is not guaranteed.
• Billing: starts at $99/month including 5 provisioned mailboxes, then $10/month for each additional mailbox ($49 platform fee + $10/mailbox, minimum 5, tiers from 5–60 mailboxes). No sending quota; sending volume is not billed; conservative planning capacity is about ≈3,300 emails per month for 5 mailboxes after warm-up. After adding mailboxes, a Stripe-hosted payment link can be obtained via POST /checkout; promo codes are entered on the Stripe checkout page.
• After payment, real mailbox provisioning is fully self-service and automatic, but mailbox sending authorization still needs to be completed on the platform side.
• Repository contents (CLI, MCP configuration, skills, plugins, integrations) use the MIT license; the hosted platform (API, engine, dashboard) is closed source, is not covered by the license, and is operated by EpiphanyMade.
• For detailed build status and service availability, refer to the official status page; missing information is not yet available.