Tray.ai review: quote-only Pro, Team and Enterprise plans, Tasks billing, the $150K AWS list price, Merlin Agent Builder, and when Workato or n8n fits better.
Tray.ai (Tray.io until the July 2024 rebrand) sells itself as an "AI Orchestration" platform: an enterprise iPaaS with 700+ connectors, a no-code agent builder called Merlin, and a gateway that publishes your workflows as MCP tools. What it does not sell is a price. Checked on tray.ai/pricing in September 2026, all three plans - Pro, Team and Enterprise - end in a sales form, and the page says pricing "is based on your usage and the features you need." Even the free trial goes through sales, with no stated length.
Two numbers leak out anyway. AWS Marketplace lists Tray Enterprise Edition, an "Enterprise License with task consumption (includes Enterprise Support)", at $150,000 for a 12-month contract, with no refunds during the term. That is a list price, and list prices at this level are opening bids. Vendr, which tracks 108 Tray purchases, puts the median at $37,641 a year, with a range of $18,700 to $96,300. Both are third-party figures, not Tray's.
The more useful Vendr number is the discount. Its buyers get about 26% off Tray's first quote, against roughly 10.6% for Boomi, whose median ($45,147) is higher. Tray's opening number has more air in it than its rivals', and one Vendr community note reports a 50% discount on Enterprise. The same notes mention a 7% renewal uplift. Cap it in the contract. At renewal it is too late.
Everything on Tray is metered in Tasks, and the documentation's own definition is the one to remember: "imagine it as equivalent to a step in a workflow." A ten-step workflow that runs 1,000 times is about 10,000 Tasks. Connector retries are free. Every Call Connector API request costs one, and so does every step you test in isolation while building - so a long debugging session shows up on the bill.
AI steps are where the maths goes hazy. A model call is measured in tokens that Tray converts into Tasks, and the ratio is not published: "Token-to-task conversion ratios are available from customer support," says the billing doc, last updated January 2026. So the workload Tray now markets hardest - agents that loop through several model calls per request - is the one you cannot price from outside. Get the ratio in writing, for the models you plan to use, before you size a Task commitment.
Once you are in, the Usage dashboard does break consumption down by workflow, so a polling trigger that fires every minute for nothing is easy to spot.
With no prices on the page, the plans differ by workspaces, how far back you can see, compliance and support. This is what tray.ai/pricing and Tray's support SLA (updated January 1, 2026) listed in September 2026.
| Pro | Team | Enterprise | |
|---|---|---|---|
| Workspaces | 3 | 20 | Unlimited |
| Insights history | 7 days | 30 days | 180 days |
| Log retention | 7 days | 7 days, expandable; log masking | Log streaming |
| SSO, HIPAA, regional hosting | Not listed | Paid add-ons | Included |
| Data engineering | Not listed | Paid add-on | Included |
| Support | Business hours, 4-hour Sev1 target | 24/7, 1-hour Sev1, Slack channel | 24/7, 1-hour Sev1, Slack channel |
Every plan includes the AI Palette, VectorTables, Tray Headless, Headless MCP, full API access and plugins for Claude Code and Codex, so developers can build Tray workflows from a coding agent.
The row that bites is Pro's. Seven days of logs means a failed sync noticed on day eight has no trail left to debug, and 30-day retention is sold separately. Support is business hours, Monday to Friday, so a Friday-evening failure in a billing workflow waits for Monday unless it counts as Severity 1 - and then the target is four hours, not one. Tray's own security statement cites a 30-day default retention, which contradicts the pricing page; ask which figure your contract carries. Downgrading is not self-serve either. Upgrades happen any time, downgrades go through support.
The "AI" in the name rests on two products. Merlin Agent Builder, launched in December 2024, is a no-code builder for conversational or autonomous agents that act through Tray workflows, with knowledge sources, governance and guardrails attached. Agent Gateway for MCP, announced in October 2025, works the other way round: IT builds governed MCP servers on Tray and publishes them, so the 700+ connectors become tools any MCP client can call through one audited door.
That second idea is the strongest thing Tray has. A company wiring agents into Salesforce and NetSuite otherwise ends up with a dozen MCP servers written by whoever needed one, each holding its own credentials. Tray already holds the credentials, the logs and the approval flow, which makes it a sensible place to put that control.
The catch is commercial. Both products sit in a paid "Agent Development" add-on that no plan includes, and Merlin's own docs route access through "your Customer Success Manager or Account Executive". The June 2025 release names OpenAI, Gemini, Bedrock and Azure as model providers; the product page says "every major LLM is pluggable", but we found no statement naming Anthropic's models or saying whether bringing your own API key avoids the token-to-Task conversion. Put both questions to sales on the first call.
Compliance is where Tray earns its enterprise pricing. The trust page lists SOC 1 and SOC 2 Type 2 reports (examined by Schellman for August 2024 to July 2025), GDPR, CCPA and HIPAA, hosting in the US, EU or APAC, and a statement that Tray never trains on your data. The AWS contract carries a 99.9% uptime SLA. The "100%" execution uptime on the About page is marketing; 99.9% is the figure you can claim against.
Leaving is harder. Projects and workflows export as JSON, but in Tray's own format, which neither Workato nor n8n imports. The export is also deliberately partial: data table rows, vector data, API management policies and agent subproject links stay behind. Tray's docs call an exported project "a portable definition of how a project is built, not a copy of its runtime content." Moving 200 workflows off Tray means rebuilding 200 workflows.
User reviews point the same way. Tray scores about 4.5 out of 5 on G2 across 157 reviews, but the recurring complaints there are the learning curve on advanced features, hard debugging and cost, with "Complex Pricing" and "Expensive" both among the tagged cons. The Tray Sync CLI, released in August 2026, adds pull, promote and drift checks between dev and prod workspaces - a real answer for teams that want integrations under version control, if not for the pricing complaint.
Gartner named Tray a Visionary, not a Leader, in its 2026 iPaaS Magic Quadrant in March 2026, the third year running in that box. If you are shortlisting enterprise iPaaS vendors anyway, Workato is the direct rival, with the same quote-only model and a longer track record in IT-owned automation. If your volume is modest and you have an engineer, n8n self-hosts on your own server and keeps workflows in a format you control. Teams that only need SaaS-to-SaaS automation without a procurement cycle should start with Make or Zapier, which publish their prices and take a card the same day. Our workflow automation roundup lines them up side by side.
Tray.ai is a capable enterprise integration platform with a credible plan for agents: govern the tools centrally, publish them over MCP, bill everything in one unit. Buy it if you are a mid-size or larger company whose IT team owns integrations, you need SOC 2, HIPAA or EU hosting, and you have five figures a year to spend - and negotiate, because Vendr's data says the first quote gives way by about a quarter.
Skip it if you need a number before a sales cycle, if you want agents without paying for an add-on on top of the platform, or if you may need to leave cleanly one day. Pin down the token-to-Task ratio before you sign. Without it, the AI half of "AI orchestration" is a cost you cannot forecast.
Visit the official website to see the latest plans, pricing, and special offers.
Get Started - Tray.ai* Affiliate link - we may earn a commission at no extra cost to you.