Designing Developer Tools for AI Agents: How Resend Captured Claude Code Recommendations

To streamline this process, Resend introduced clean, machine-readable alternatives, such as appending .md to pricing URLs to serve structured Markdown tables directly to agents in a single request.

Resend's strategy also builds upon earlier distribution efforts, such as React Email. Before launching its primary sending API, the team published React Email to simplify email template creation. Developers could design templates with React Email and send them using alternative providers like SendGrid. As a result, Resend's code base was already present in many software repositories before agents or developers evaluated backend migration.

MCP server design: Tool expansion versus context limits

As protocol standards evolve, integrating with Model Context Protocol (MCP) servers has become a core mechanism for enabling agent actions. At launch, Resend’s MCP server offered roughly 20 tools or fewer. It has since expanded to more than 80 distinct tools, aiming to replicate the full functionality of its web dashboard—including sending messages, inviting team members, checking resource usage, and managing account billing.

This comprehensive tooling approach contrasts with alternative designs, such as Cloudflare’s Code Mode server, which exposes its API using just two core functions: search() and execute(). Expanding tool counts presents distinct design trade-offs. Rocha noted that inspecting agent execution logs reveals instances where response payloads exceed the model's context window. Merely reducing the number of exposed tools does not automatically reduce the size of the returned context, prompting developers to carefully optimize result payloads for daily tool calls.

What it means for developers

For software engineers and platform builders, the rise of agentic coding tools changes how developer products are evaluated and integrated:

• Machine-friendly interfaces: Adding plain Markdown endpoints (.md) and predictable API responses ensures that AI agents can evaluate pricing, schemas, and docs without parsing complex client-side rendering.
• Context window efficiency: Building MCP servers requires monitoring how much data tool calls return, ensuring agent prompts are not overwhelmed by oversized responses.
• Distribution through open components: Establishing utility libraries early—such as template engines—can place codebase dependencies before an agent even selects an API provider.

As developers test how different AI models make architectural decisions or handle agentic workflows, accessing multiple LLMs efficiently becomes crucial. Developers can try top AI models cheaply through one API at https://apixoai.online to evaluate prompt behavior across models like Claude, GPT, and DeepSeek.

Infrastructure, deliverability, and anti-abuse efforts

Winning an agent's initial selection via code generation is only the first step; delivering infrastructure reliably remains a core technical challenge. Email delivery systems frequently face security risks associated with shared tracking domains. If a bad actor uses a shared domain for phishing, legitimate transactional emails across other accounts can suffer from blocklisting.

To mitigate these risks while preserving developer access, Resend introduced custom tracking domains alongside a $3M safety commitment to curb spam on its free tier. According to Rocha, Resend had grown to 4.3 million users and 120,000 paying customers at the time of recording, highlighting the scale required to maintain deliverability while serving both human developers and autonomous AI agents.

Source: Designing for Agents — Zeno Rocha, Resend — Latent Space. Written by the Apixo team from that report.