What it is
Tray.ai — Tray.io until its rebrand — is a low-code integration and automation platform from San Francisco, co-founded in 2012 by Rich Waldron, who is still CEO. It sits in the middle of the market: more governance and depth than Zapier or Make, a lower floor and a lighter implementation than Workato, and none of the self-hosting work that n8n asks of you.
Gartner placed it a Visionary in the Magic Quadrant for iPaaS published on 18 March 2026, its third consecutive year there, and one of four vendors in that quadrant. Be precise about what that means: Visionary is a completeness-of-vision placement, not an ability-to-execute one. Workato and Boomi are Leaders in the same report. Public customer references include Life360, Apollo, Yext and Zuora.
What changed in 2026
Three things make Tray a different product from the one you may have evaluated in 2023.
Tray Headless, released on 3 June 2026, puts the platform inside an AI coding tool. It ships two ways: a tray-workflows plugin for Claude Code, and a hosted MCP server at api.tray.io/mcp that Cursor, Codex, Windsurf or your own agent can drive. You describe an integration in plain language — “sync new Salesforce leads to Slack” — and it plans the workflow, picks connectors, sets up authentication, validates the steps and test-fires them with your approval. Anything built this way opens in the visual builder with no migration, so an engineer can draft in text and an ops owner can maintain it on the canvas. Authentication is OAuth sign-in in the browser, not a pasted API token. The catch at launch: Headless is limited to US-region workspaces.
Agent Gateway is the other direction — publishing your Tray workflows and 700-plus connectors as governed MCP tools that external agents can call, with access control, authentication options and observability. Merlin Agent Builder is Tray’s own agent layer, with guardrails, audit logs, knowledge ingestion, memory, multi-LLM support and deployment to Slack and Teams, plus prebuilt ITSM and HR agents. Both sit in the paid Agent Development add-on; Headless MCP, the Claude Code and Codex plugins, and full API access are included in every plan.
A use case worth buying it for
Lead-to-account routing that has outgrown a Zapier account. A form fill or enrichment event lands, Tray matches it to an existing account in Salesforce or HubSpot, applies territory and round-robin rules, creates or updates the record, and posts the assignment to the owner in Slack — with branching for duplicates, a named error path when the CRM rejects the write, and logs someone can read afterwards. This is the shape Tray handles well: multi-branch logic across two or three systems of record, owned by a RevOps or systems team rather than an integration centre of excellence. Under Headless, a RevOps engineer can draft that workflow from Claude Code in an afternoon and hand it to an admin who maintains it visually.
Pricing
Tray publishes three plans — Pro, Team and Enterprise — and no prices for any of them. Pro gives 3 workspaces and 7-day log retention; Team gives 20 workspaces, 30-day insights and log masking; Enterprise adds unlimited workspaces, 180-day insights, SSO, HIPAA, regional hosting and on-premise capability. All three draw on one metered unit, the task.
Vendr’s transaction data is the anchor: a median of $37,641 per year across 108 purchases, a range of roughly $18,700 to $96,300, and an average 26% off first quote. Vendr puts Boomi’s median at $45,147 by comparison, and Workato’s published median sits around $64,500 — so Tray usually lands at roughly half to two-thirds of a Workato contract for comparable scope.
How tasks count: one workflow step equals one task, however many API calls that step makes internally. Every connector execution inside a loop is a task, so a three-step loop over 500 records is at least 1,500 tasks. Retries are not charged. AI token usage converts to tasks at a customer-specific ratio you have to ask Customer Success for. Tray does not stop your account when you exceed your allocation; usage alerts fire at 80% and 100%, and overage is billed at the on-demand task rate afterwards.
Best for
RevOps, systems and business-technology teams at 200-to-2,000-person companies that need governed, multi-system automation with a real error path, have one or two technical owners rather than a platform team, and want agent and MCP tooling without a Workato-sized contract.
Alternatives, and when to pick them instead
- Workato — a Gartner Leader for the eighth time in 2026. Pick it when a named integration team owns ERP-grade processes such as order-to-cash, you need formal environment promotion, and the budget is already six figures.
- Boomi — a Leader for the 12th consecutive time and highest for Ability to Execute in the 2026 report. Pick it when the estate is data and application integration first and the longest track record matters more than the agent layer.
- n8n — the fastest-growing entrant, valued at $5.2B in May 2026. Pick it when you can self-host, have engineers, and want no floor on price.
- Zapier — pick it when the workflows are linear app-to-app moves owned by non-technical people and nobody will read the logs.
Watch-outs
- Loops make task forecasts wrong by an order of magnitude. Guard: before signing, take your five highest-volume planned workflows, count connector executions per run including every loop iteration, multiply by expected monthly runs, and negotiate the task pool against that number. Ask Customer Success for your token-to-task ratio in writing if any workflow calls an LLM.
- Overage is billed after the fact at the on-demand rate, with no hard stop. Guard: get the on-demand task rate into the order form, and wire the 80% usage alert to an owner who can pause a runaway workflow the same day.
- Price escalators at renewal. Vendr buyers report a standard 7% annual uplift. Guard: trade a multi-year term for a fixed rate, and cap the renewal increase in the contract rather than negotiating it in month eleven.
- Headless is US-region only today. Guard: if your workspace is hosted in the EU or another region for data-residency reasons, treat Headless as unavailable and evaluate on the visual builder alone.