What is Relay.app doing strategically?

Relay launched a dedicated Agents product on February 3, 2026, anchoring the homepage and navigation around building teams of AI agents rather than building workflows. The company page and product tour now lead with agent language, and the AI Agent Academy was added as a retention and education surface.

This is a positioning bet, not a feature launch. Buyers searching for AI agent builders now see Relay in that category rather than the Zapier-alternative bucket. It compresses the evaluation shortlist and changes who Relay gets compared against in procurement.

The repositioning is coherent and backed by product, pricing, and content all pointing the same direction. The risk is that calling yourself an AI team builder raises the bar on what buyers expect from autonomous execution. Relay's human-in-the-loop controls are a real differentiator, but competitors who move faster on full autonomy could outpace the narrative.

All Relay plans include every integration, with usage measured in steps and AI credits rather than per-app paywalls. The Team plan at $138 per month for up to ten users, with a 50 percent annual discount, creates a predictable budget line for small ops teams. Trigger polling does not consume steps, which reduces sticker shock during evaluation.

Zapier's per-task billing stacks up fast at volume and premium app integrations sit behind higher tiers. Relay's structure removes both objections in a single comparison. For a ten-person ops team running moderate automation volume, the annual math often favors Relay by a significant margin.

The pricing architecture is a genuine competitive weapon in mid-market sales cycles. The ceiling is real: 100,000 steps per month costs approximately $1,100 on top of plan fees, which limits Relay's value story at enterprise scale. But below that threshold, the model is clean and easy to defend in a budget conversation.

Relay introduced Model Context Protocol support in July 2025, allowing users to create MCP servers inside Relay and expose workflow-driven tools to external AI environments like Claude Desktop and Cursor. The reverse is also supported: Relay agents can call external MCP servers as tool sources within a workflow.

MCP is rapidly becoming the connective tissue between AI applications and business systems. A workflow platform that can act as an MCP server gains reach beyond its native integration list. Relay users who build MCP servers are embedding Relay deeper into their AI stack, which raises switching costs materially.

This is the right move at the right time, but the feature is still marked as advanced and is currently limited to Claude models for remote MCP connections. That constraint keeps it out of broad adoption for now. Watch whether Relay extends MCP support to OpenAI and Gemini in Q2 to Q3 2026 as the signal for whether this becomes a mainstream growth driver.