Retool

What it is

Retool is an internal-tool builder. You point it at Postgres, Snowflake, a REST API or Salesforce, drag components onto a canvas, and write the glue in SQL and JavaScript. It has been the default answer to “we need an admin panel by Friday” for years, and it runs at Amazon, DoorDash, Brex, Rippling, NBC and Plaid. Retool reported $120M ARR in October 2025, up from $90M at the end of 2024, on $190M raised at a $3.2B valuation.

What changed in 2025-26 is not the canvas. It is the meter.

Three products, three units of billing

Retool now sells three layers and prices each one on its own unit. That is the reason to read the pricing page instead of skimming it.

• Apps bill per seat, split by what the person does: builders who edit, internal users who click, external users on a volume curve.
• Workflows bill per run — 500/month free, 5,000 on both Team and Business, then $75 per additional 5,000 runs.
• Agents bill per agent-hour. Every organization gets $50 of agent usage a month, which Retool puts at up to 20 hours depending on the model. Past that you pay by the second, itemized per model on the invoice.

Retool argues the case out loud: per-seat pricing punishes you as more people use the thing, token pricing hides the number until the invoice lands, and a flat $10K/month costs the same for 100 tasks as for 10,000. Its own comparison anchor is a mid-level analyst at “$50-80 per hour after benefits, overhead, and management” — vendor framing, but it makes Retool the first mainstream ops tool where finance can set agent spend beside a headcount line in the same unit.

The practical consequence is three cost curves on three different axes. You can put a read-only console in front of 200 internal users at $15/seat, run the automation behind it by the run, and pay for the drafting agent by the hour, instead of forcing one seat count to absorb all three.

Retool also shipped its own MCP server on 8 May 2026 — public beta, OAuth 2.0 over HTTP, available on all plans and on self-hosted. Claude, Cursor, ChatGPT or Codex connect to your workspace and build apps, edit workflows, run queries against connected resources and manage users, scoped by the permissions you authorize.

A use case worth buying it for

The refund-and-credit approval console for support ops. Retool reads billing history from Stripe and the warehouse into one view, an agent drafts the customer-facing note from the ticket thread, and anything above your approval threshold waits behind a button a named human presses. Write-back goes to Zendesk and the billing system in a fixed order. What you get that a spreadsheet-and-Slack process does not give you is one record per credit issued naming the approver — and the agent hour is charged only while the draft is being written, not while it sits waiting for that approval.

Pricing

Free is $0 with 500 workflow runs and 250 AI credits a month. Team is $10 per builder and $5 per internal user monthly on annual billing ($12 and $7 month-to-month), with 1,000 AI credits. Business is $50 per builder and $15 per internal user on annual ($65 and $18 monthly) with 3,000 AI credits; external users are free to 50, then $8/user to 250, $6 to 500 and $4 above that. Enterprise is quote-only at 10,000 credits.

The seat price is not the number you will pay. Vendr’s February 2026 report on 156 tracked Retool purchases puts the median contract at $64,623/year, with the observed range running $24,500 to $152,500 and buyers landing 18.6% below first quote on average. Budget $25K/year as the floor for a real deployment.

Best for

RevOps, support ops and CS ops leaders at 100-1,000-person companies that already run a warehouse or Postgres and have 2-6 people who write SQL comfortably. Retool’s speed comes from those people; the drag-and-drop is the last 20% of the job.

Do not buy it if nobody on the team writes SQL and JavaScript — you will pay a contractor to maintain the app and the seat price stops being the cost. Do not buy it if the internal tool you have in mind is a SaaS-to-SaaS automation with no interface in front of it. That is n8n, Zapier or Tines, and all three are cheaper for that job.

Alternatives, and when to pick them instead

• Microsoft Power Apps — the enterprise incumbent by installed base, because it arrives inside the M365 licensing your firm already pays for. Pick it when the app is a form over SharePoint or Dataverse and the buyer is IT rather than engineering. The bundle economics beat Retool’s per-builder price; the developer experience does not.
• Appsmith — Apache 2.0, self-hosted, unlimited users on the community edition, 35K+ GitHub stars and $51.5M raised. Pick it when per-seat cost at 200+ internal users or data residency is the blocker and you have infrastructure people to run it. You give up the agent layer entirely.
• Superblocks — the AI-generation entrant, $60M raised, whose Clark agent generates internal apps from a prompt under governance rules. Pick it when the goal is non-engineers producing apps that IT still controls, rather than engineers building faster.

Watch-outs

• 5,000 workflow runs is the easiest ceiling to hit, and Business does not raise it. One cron on a 5-minute interval is 8,640 runs a month by itself — past the allowance before a single person clicks anything. Guard: count triggers times frequency before you migrate, and where a schedule is the trigger, widen the interval or batch the payload. At $75 per 5,000 runs the arithmetic moves fast.
• The agent meter bills failed runs. Billable time runs from trigger to answer whether or not the agent succeeded; only waits on subagents, resource authentication and human approval are excluded. A retry-heavy agent on an expensive model burns the $50 allowance without producing output. Guard: run one agent for a full billing month inside the free allowance, read the per-model line items on the invoice, then pin the cheapest model that clears your accuracy bar before you enable a second agent.
• Agent overages are invoiced monthly even on an annual contract. Guard: treat agent spend as opex with its own monthly ceiling rather than folding it into the annual line, and set the review on the first two invoices — the seconds-level per-model detail is where you find the agent nobody meant to leave running.
• Above Business the price is a negotiation and your first quote is a starting position. Guard: anchor on Vendr’s published band instead of the seat math, and get a Superblocks or Power Apps quote in the file before you counter. The 18.6% average saving is not what you get by accepting the number you are handed.