Manifest Review (2026): Pricing, Features & Honest Verdict

Manifest is an API that turns any webpage into a structured JSON list of the actions an AI agent can take on it. It solves a real gap between browser tools and scrapers, but it meters you on a unit it never defines. Best for: developers building web-acting agents. Price: from $29/mo (free plan: yes, 50 calls). Rating: 7/10.

What is Manifest?

Manifest sits between two things that already exist and neither of which does this job. Give an agent a headless browser and it gets a DOM. Give it a scraper and it gets text. Neither tells the agent what it is allowed to do on the page: which buttons exist, which fields are required, and what has to happen before the submit button will work.

You POST a URL. Manifest visits it, reads the accessibility tree the way a screen reader does, cross-references that against the actual DOM for the things accessibility trees miss, and returns a JSON action manifest. Every button, form and input, each with a resolved locator giving both a CSS selector and an ARIA role plus accessible name, and a requires field listing the ids of other actions that must be completed first.

We found it at number 11 on the 21 July Product Hunt board. It is built by Max Nordström, a solo founder operating as Omfang AB in Sweden, and the product name and the company name are different, which is worth knowing before you go looking: the site is omfang.io and the API lives at manifest.omfang.io.

What Are Manifest’s Key Features?

The requires field

This is the part that justifies the product. A returned action carries a requires array naming the other actions it depends on. A submit button will list the email input and password input ids, so an agent knows the button is blocked rather than discovering it by clicking and failing. Most agent frameworks handle this by trying things and reading errors. Manifest hands the dependency graph over up front.

Dual locators that survive redesigns

Every action comes back with both a CSS selector and an ARIA role plus accessible name. The docs explicitly tell you to prefer the role and name because they hold up better when a page ships a redesign. That is correct advice and it is unusual for a vendor to write it down, since the fragile-selector problem is exactly what they are selling against.

MCP server as the default integration

The preferred integration is not the REST API, it is an MCP server at io.omfang/manifest, exposed as a streamable HTTP endpoint and listed on Smithery. REST is offered as the fallback for tools that do not speak MCP yet. Leading with MCP rather than bolting it on later is a good read of where agent tooling is going.

A copy-paste system prompt

The homepage carries a block of text you paste into your agent’s system prompt to teach it when to reach for Manifest and how to read the response. It explains the positioning in the agent’s own terms: a browser gives you a DOM not a list of things you can do, content extraction gives you text not actions. Small touch, and it removes the most annoying part of adopting a new agent tool.

Response caching

Cached responses come back fast and, per the pricing note, do not count against your rate limit the same way a cold request does. Most agents hit the same handful of URLs repeatedly, so this matters more than it sounds. The wording is loose about how cache hits interact with your call quota, which we cover below.

What is a fingerprint call?

We do not know, and neither will you.

Manifest’s pricing table meters two separate units. Manifest calls per month, which is clear enough. And fingerprint calls per month, which appear in every tier at five times the manifest allowance: 250 on Free, 5,000 on Starter, 25,000 on Pro.

The phrase “fingerprint” appears on omfang.io exactly where the pricing table puts it and nowhere else. There is no definition in the feature copy, no mention in the how-it-works section, and nothing in the pasted agent instructions. You are being metered on a unit the marketing site never explains.

Our best guess is that it is a lighter-weight endpoint that identifies a page rather than mapping it, which would explain the 5:1 ratio. But a guess is what it is. If you are sizing a plan around this product, the fingerprint allowance is one of your two constraints and you cannot currently reason about it without asking. Check the docs or email them before you commit.

How much does the overage actually cost?

More than the plan, by roughly three times, and the cliff is steep enough to matter.

Starter is $29/mo for 1,000 manifest calls, which is $0.029 per included call. Overage is $0.08 per call, so every call past the allowance costs 2.8 times what the ones inside it cost. Pro is $79/mo for 5,000 calls, which is $0.0158 per included call, against $0.05 overage, a ratio of 3.2.

Run the cliff. A Starter customer who makes 5,000 calls in a month pays $29 plus 4,000 overage calls at $0.08, which is $349. The same 5,000 calls on Pro cost $79. You are paying 4.4 times too much for not noticing you outgrew your tier.

This is not unusual pricing and Manifest is not hiding it, both overage rates are printed in the table. But there is no published usage alert, no auto-upgrade, and no soft cap mentioned anywhere, so the discipline is entirely on you. Set your own monitoring on call volume before you put this in a production agent loop.

One smaller thing, in the same table: the fourth tier is spelled “Enterprice”. It is a typo on a solo founder’s marketing page rather than a product defect, and we mention it only because it is the kind of polish gap that tells you how young this is.

How Much Does Manifest Cost?

Manifest starts free and the first paid tier is $29 a month. Free gives you 50 manifest calls and 250 fingerprint calls per month with no overage available. Starter at $29/mo gives 1,000 manifest calls and 5,000 fingerprint calls, with overage at $0.08 per call. Pro at $79/mo gives 5,000 manifest calls and 25,000 fingerprint calls, with overage at $0.05. Enterprise is custom.

The pricing lives as a section on the homepage. The /pricing route in the nav returns an error page, so if you land there directly you will think the product has no published prices. Scroll the homepage instead.

Compared to adjacent infrastructure, $29 is modest. Browserbase, which gives agents actual cloud browsers rather than action maps, publishes a free tier, Developer at $20/mo and Startup at $99/mo with custom Scale plans above that. Firecrawl sells usage-based credits with a free tier and custom enterprise plans, though its tier figures render client-side and we could not confirm them on the vendor’s own page, so we are not quoting numbers. Steel.dev is open-source browser infrastructure for agents, which makes self-hosting the price floor if you want to run this class of thing yourself.

Note that these are not substitutes. You will likely pay Browserbase or Steel for the browser and Manifest for the action layer on top, so read $29 as an addition to your agent stack cost, not a replacement for anything in it.

Who is Manifest Best For?

Use Manifest if you are building an agent that fills forms, completes checkouts, or navigates multi-step flows on sites you do not control, and you are tired of hand-maintaining selectors that break on every redesign. The requires graph is the specific reason to pick it: knowing that submit is blocked on two empty fields is information your agent otherwise has to learn by failing.

Use it if you have already adopted MCP. The integration is one config entry and the vendor clearly expects that to be the main path.

Skip Manifest if your agent only reads pages. If you need text and not actions, a scraper is cheaper and simpler and this adds a hop for nothing. Skip it if you are operating at a volume where the overage cliff is a real financial risk and you cannot put monitoring around it. And think carefully if you need vendor stability: this is one person and one small Swedish company, with a free tier of 50 calls and a pricing table that has a typo in it. Write your integration behind an interface you can swap out.

Best Manifest Alternatives

Browserbase gives your agent a real managed cloud browser rather than a map of what it can click. Free tier, Developer at $20/mo, Startup at $99/mo, custom Scale above. It is the layer underneath Manifest more than a competitor to it, but if your agent framework already drives a browser well, you may not need the action map.

Firecrawl handles search, scrape and crawl with usage-based credits, a free tier and custom enterprise plans. Reach for it when you need page content at scale. It will not tell you what is clickable or what is required.

Steel.dev is open-source browser infrastructure for AI agents, with scalable cloud sessions behind a developer API. The open-source path is the real alternative here: if the fingerprint-unit ambiguity or the solo-founder risk bothers you, running your own browser layer and writing your own accessibility-tree extraction is a weekend of work rather than a quarter.

Final Verdict: Is Manifest Worth It?

The idea is right and the execution of the core object is better than we expected. Reading the accessibility tree instead of screenshotting the page is the correct approach, returning both a CSS selector and an ARIA role plus name is the correct hedge, and the requires field is a good piece of API design that we have not seen packaged this cleanly elsewhere. At $29 a month it is a cheap thing to try, and the free tier is enough to prove whether the JSON is useful for your agent.

What holds it back is everything around the product rather than the product. You are billed against a unit the site never defines, the pricing route in the nav is broken, the top tier is misspelled, and overage runs at roughly three times the in-plan rate with no published alerting. None of that is fatal and all of it is fixable in an afternoon. Try it, keep your integration swappable, and email them about fingerprint calls before you pick a tier.