Taste Labs hands your agents a brand’s look. Jinn Agents hands them the whole verified record: identity, voice, claims, and design, over one token.
Taste Labs is a serious piece of engineering aimed at one layer of brand: the look. Give its Brand API a URL and it returns a structured design system, colors, typography, layout, surfaces, components, and a brand profile, plus the HTML, CSS, and screenshots it was built from; describe a mood and Style Search returns ranked brands from a curated corpus; give it two URLs and the Verifier scores how well one page follows the other’s visual brand, with worst-first fixes an agent can apply in a loop. Three subsystems, one API, priced per credit from $99 a month with 20 free credits to start. What it does not hold is the rest of your brand. The profile is extracted, not verified; the voice is a handful of tone adjectives; we found no place in its docs (read 2026-10-10) to confirm a claim, ban a word, or correct a fact, and nothing checks what an AI actually tells your buyers. Jinn Agents returns that whole record over one bearer token: identity, positioning, voice with banned words, messaging pillars, and design tokens, extracted from your site and verified by you. Every call is checked against who is asking and what their plan allows, and the read an outside agent gets leaves your competitive intel out. If your agents only need to match a look, Taste Labs does that one job well. If they need to be your brand, start with the free Jinn demo token.
Taste Labs competes with Jinn Agents, the MCP and agent-infrastructure arm of Jinn. The rest of the suite (Fama, Ghost, Vermeer, Chart, Silver) is not part of this comparison.
Updated October 10, 2026
A brand-grounded MCP gateway: one bearer token, and tools that return a verified Brand DNA record extracted from your site so any AI answers as your brand. One arm of the Jinn suite; a free demo token is the front door.
A Brand API for agents with three subsystems: Brand Extraction (beta) returns a site’s visual design system, Style Search (alpha) returns ranked brands from a curated corpus, and Verify Adherence (alpha) scores a page against a reference site. Credit-priced, with a REST API and a hosted MCP server.
Where the two actually differ
Contextual rows, not checkmarks. Every Taste Labs cell cites the Taste Labs page it was verified on, with the date. Every Jinn cell quotes the same locked facts the Agents product pages ship.
| Jinn Agents | Taste Labs | |
|---|---|---|
| What it is | A brand-grounded MCP gateway: one bearer token and a set of tools that return your verified Brand DNA so agents answer as your brand, not as a model average. | In its own words: make your agents be on brand. Extract any brand system, search for style inspiration, and verify if it is staying on course. One API, three subsystems, all about the visual layer.[1] |
| What a tool call returns | The verified extracted record: identity, positioning, voice with banned words, messaging pillars, design tokens. The whole brand, not one layer of it. | A structured design system: colors, typography, layout, surfaces, components, and a brand profile, plus the captured HTML, CSS, screenshots, and assets it was built from. Detailed enough to replicate a site.[3] |
| Verified, or extracted | The record is extracted from your site and verified by you before agents read it. The agent writes from returned facts, not from a style guess. | Extracted automatically from the URL you submit and cached per URL, so a URL someone already extracted comes back immediately. The docs call the design system what the brand declares; no human verification step is described.[2] |
| Voice, claims, and banned words | Voice register, palette, positioning, and claims come back together from one call, so an agent that writes copy and an agent that builds a page read the same record. | The brand profile carries a copy_tone field of tone adjectives such as Confident, Technical, Direct. There are no claims, messaging pillars, or banned words in the documented schema, and no field a human corrects.[7] |
| What gets checked | The record itself is the checked thing: it is the same verified Brand DNA the Jinn products run on, including the arm that audits what AI engines say about you. | Visual adherence: give it a reference site and a page to judge, and the Verifier returns one score from 0 to 1, prose recommendations, and structured fixes with exact target values, worst-first and capped at 20. Whether an AI tells buyers the truth about you is out of scope.[4] |
| Where a person steps in | Two tools write, and neither can spend: one hands back a human-action URL, the other files a proposal a person approves. An agent prepares; a person completes. | The agent-loop guide feeds the verdict back to the agent that built the page, caps the loop at two or three passes, and escalates to a human beyond it. The person is the fallback after the loop, not a gate inside it.[5] |
| Access and gates | Bearer tokens, checked on every call against three independent gates: audience, scope, and tier. A paid subscription never widens an agent to operator-level trust. | A hosted MCP server at mcp.tastelabs.com/mcp with OAuth sign-in or a taste_ API key sent as a bearer token; the same key authenticates the HTTP API. Works in any MCP client.[6] |
| What leaves the boundary | What an outside agent sees is a bounded public projection; competitive intel is never included in the public read, and the audit trail redacts free text before hashing. | Style Search does not read your submissions; it searches a prebuilt corpus of curated brands with palette, typography, and taxonomy metadata. Your own extractions are cached per URL on Taste Labs’ side.[2] |
| How usage is metered | Tier is resolved from the live subscription on every call, fail-closed. Nothing to re-mint when billing changes, and no per-call credit math. | Credits per call: Search is 1 credit light or 2 deep, Extraction 2, Verifier 2. Every extraction call spends credits even when served from cache; the docs put it plainly: a cache hit is faster, not free. Failed runs are refunded.[1] |
| Maturity labels | Early access, with the gates already in place: the token is brand-scoped and the tier is checked live on every call. | Labeled on its own page: Brand Extraction is beta, Style Search and Verify Adherence are alpha. The home page says it works with the top frontier labs.[1] |
| Pricing shape | Free demo token first, then Connected $99, Brand $449, and Agency $1,499 a month. Each Jinn product is priced on its own. | Credit bundles: Starter $99 a month for 600 credits, Pro $749 a month for 6,000, top-ups at $99 per 600, Enterprise on request, and 15% off on annual billing. 20 free credits to start.[1] |
| One piece or the whole system | One arm of a connected system: the record Agents returns is the record Ghost writes from, Vermeer designs from, and Fama audits AI answers against. | One piece, done deliberately: extraction, search, and adherence all operate on the visual design system. Copy, claims, strategy, and what AI says about you are someone else’s job.[2] |
Bracketed numbers cite the dated Taste Labs sources at the foot of this page. Nothing here relies on a review-site or aggregator figure.
One piece, or the whole system
Taste Labs hands an agent the look of a brand, and it is honest about that scope: colors, typography, layout, components, and a verdict on visual adherence. A brand is also what it claims, how it speaks, which words it never uses, and what buyers are told about it. Jinn Agents returns all of that from one call, verified by you. This whole page is about whether your agents need to match a look or be a brand.
Visual adherence is not brand truth
The Verifier answers one question: does this page follow that site’s visual brand? A page can score well and still state the wrong price, invent a feature, or misquote your founders. The record Jinn Agents returns is the same verified Brand DNA the Fama arm audits AI answers against, so the facts an agent writes from are the facts you confirmed.
Extracted and cached, or extracted and verified
Taste Labs extracts from the URL you submit and caches the result per URL; its docs describe the design system as what the brand declares. Jinn Agents also starts from your site, then puts a person in the loop: you verify the record before agents read it, and corrections stay in it. When agents write as you, who checked the record is the part to ask about.
One arm of a connected record
The Brand DNA record Agents reads is the same record Ghost writes from and Vermeer designs from. Grounding your agents and staying consistent everywhere else are one motion, not a design API plus a set of point tools.
What it costs, and what sits underneath
The grid names the pricing shape; this is the detail. Every Taste Labs figure was read off tastelabs.com/api on 2026-10-10. Every Jinn line quotes the locked Agents pricing facts.
| Jinn Agents | Taste Labs | |
|---|---|---|
| How you pay | A flat monthly price per tier, with a free demo token first: Connected $99 a month, Brand $449, Agency $1,499 a month. The price is the tier, not a count of calls. | Credits: Starter $99 a month for 600 ($0.165 a credit), Pro $749 a month for 6,000 ($0.125 a credit), top-ups at $99 per 600, Enterprise on request. Annual billing takes 15% off, to $84 and $636 a month.[1] |
| Seats and access | Access is a brand-scoped bearer token, tier checked live on every call, that you rotate or revoke from your account. Revocation is immediate. | A taste_ API key from the Engine dashboard authenticates both the HTTP API and the hosted MCP server, or you sign in to the MCP with OAuth. Seat counts are not part of the published pricing.[6] |
| What each tier gates | Depth of the record. Connected grounds agents on your own brand and answers their questions with canonical facts. Brand unlocks the full typed record and the render-ready kit of colors, fonts, and logo. Agency adds the operator surface across every client brand you hold keys for. | Credits, not features: Pro includes everything from Starter at a lower per-credit rate. Each call has a fixed cost, Search 1 or 2 credits, Extraction 2, Verifier 2, and a cached extraction still spends.[1] |
| Cheapest way to see it grounded | The free demo token: no login, read-only, three showcase brands, a sixty-minute window. You watch a real call come back grounded before you pay anything. | 20 free credits on sign-up, enough for a handful of extractions or verifications at 2 credits each, then Starter at $99 a month.[1] |
Every Taste Labs figure here is Taste Labs’ own published number, read on 2026-10-10. Taste Labs publishes per-credit prices and per-call credit costs but no seat counts, so none are stated.
Jinn Agents is just the start.
Jinn Agents is one arm of Jinn, and every arm reads from the same verified Brand DNA record. The record your agents pull over MCP is the record Ghost writes from, Vermeer designs from, Chart strategizes from, and Fama audits AI answers against. You confirm your brand once, and every product you switch on next compounds that same record instead of starting from a blank prompt.
One record, every product
The verified Brand DNA that Agents returns is the exact record the rest of the suite builds on. Confirm it once and it’s right everywhere, not re-extracted per tool.
Each arm sharpens the next
Every correction you confirm and every fact you add deepens the one record, so the next product you turn on starts from everything the last one already learned.
Priced arm by arm
There’s no bundle you have to buy. Each Jinn product is priced on its own, so you start with the free demo token and add the next arm when it earns its place.
Where Taste Labs is genuinely stronger
An honest comparison names where the other side wins. Three places, all verified today on Taste Labs’ own pages.
Depth on the visual layer, plus a search corpus
Taste Labs returns a design system detailed enough to replicate a site, down to the captured HTML, CSS, screenshots, and assets, and Style Search finds ranked brands from a curated corpus when you have a mood but no URL. If the job is building pages that look exactly right, that depth is the product. Jinn Agents returns design tokens as one part of the whole record, not a site replica.[3]
A verifier built for agent loops
The Verifier returns a score from 0 to 1 with worst-first recommendations and structured fixes carrying exact target values, so an agent can apply them, redeploy, and re-verify. That is a real eval signal for generated pages, and Jinn Agents has no equivalent visual-adherence score; its check is that the record itself is verified before an agent reads it.[4]
Transparent per-credit pricing with a free start
Every call has a published credit cost, every tier a published per-credit price, failed runs are refunded, and 20 free credits need only a sign-up. You can model the bill before you run it. Jinn Agents is a flat tier instead, with a free demo token rather than a credit grant.[1]
The questions that decide it
- Is Jinn Agents a replacement for Taste Labs, or something different?
- Different scope. Taste Labs is a Brand API for the visual layer: extract a site’s design system, search a curated corpus of brand systems, and score a page’s visual adherence to a reference site. Jinn Agents is the MCP arm of Jinn: one bearer token that returns your whole verified brand record, identity, positioning, voice with banned words, messaging pillars, and design tokens, with tier and audience checked on every call. Teams building pixel-faithful pages are evaluating Taste Labs; teams that need agents to speak and act as the brand are choosing what Jinn Agents does.
- Does Taste Labs verify the brand record, or check what AI says about my brand?
- Neither, on the pages we read. Extraction is automatic and cached per URL, the profile’s voice is a list of tone adjectives, and the Verifier scores a page’s visual adherence to a reference site. Nothing checks a claim against facts you confirmed or looks at what AI engines tell buyers. The Jinn record is verified by you, and the Fama arm audits AI answers against that same record.
- Can my agents use both?
- Yes. Both expose MCP servers that work in Claude, Cursor, or any MCP client, so an agent could pull the verified brand record from Jinn Agents and score a generated page’s visual adherence with Taste Labs. They answer different questions; one does not replace the other on the visual layer.
- How do the two handle agents that act, not just read?
- Jinn Agents has two write tools and neither can spend: one hands back a human-action URL, the other files a proposal a person approves. Taste Labs’ guidance for its agent loop is to cap it at two or three passes and escalate to a human beyond that. One puts the person inside the loop as a gate; the other puts the person after it as a fallback.
- What does it cost to start?
- With Jinn Agents, nothing: a free demo token needs no login and works read-only against showcase brands for sixty minutes, then Connected is $99 a month. With Taste Labs, 20 free credits on sign-up, then Starter at $99 a month for 600 credits or Pro at $749 for 6,000, with extraction and verification at 2 credits a call.
Updated October 10, 2026
- [1] Taste Labs Brand API page (make your agents be on brand: extract any brand system, search for style inspiration, and verify if it is staying on course; 20 free credits; Brand Extraction beta, Style Search alpha, Verify Adherence alpha; Starter $99 a month for 600 credits at $0.165 a credit, Pro $749 a month for 6,000 credits at $0.125 a credit, top up at $99 per 600 credits, Enterprise, annual 15% savings at $84 and $636; Search light 1 credit or deep 2, Extraction 2, Verifier 2; results cached per URL; failed runs refunded) https://tastelabs.com/api accessed 2026-10-10
- [2] Taste Labs docs, architecture (one API, three subsystems; the design system is what the brand declares; Search does not read your submissions and searches a prebuilt corpus of curated brands with palette, typography, and taxonomy metadata; the verdict has one score from 0 to 1, prose recommendations, and structured fixes with exact target values) https://docs.tastelabs.com/concepts/architecture accessed 2026-10-10
- [3] Taste Labs docs, changelog (extraction returns a structured design system: colors, typography, layout, surfaces, components, and a brand profile, plus the captured HTML, CSS, screenshots, and assets; detailed enough to replicate a site; extraction runs as an asynchronous job) https://docs.tastelabs.com/changelog accessed 2026-10-10
- [4] Taste Labs docs, brand adherence (recommendations worst-first with exact target values, capped at 20; structured fix objects worst-first, capped at 20, each with an action discriminator) https://docs.tastelabs.com/concepts/brand-adherence accessed 2026-10-10
- [5] Taste Labs docs, agent loop (feed the verdict back to the agent that built the page and iterate; cap the loop at two or three passes and escalate to a human beyond it) https://docs.tastelabs.com/use-cases/agent-loop accessed 2026-10-10
- [6] Taste Labs docs, MCP server (endpoint mcp.tastelabs.com/mcp; connect with OAuth sign-in or send a taste_ API key as a Bearer token; the same key authenticates the HTTP API and the MCP server; works in any MCP client; every extract_brand call consumes credits even when served from cache, a cache hit is faster, not free) https://docs.tastelabs.com/ai-tools/mcp accessed 2026-10-10
- [7] Taste Labs docs, design system (the brand profile’s copy_tone field holds tone adjectives such as Confident, Technical, Direct) https://docs.tastelabs.com/concepts/design-system accessed 2026-10-10
Jinn Agents facts on this page quote the same locked content module the Jinn Agents product pages ship, checked by test.
Your agents are already speaking for your brand.
Every agent you run is describing your company to someone, in copy, in code, in a page it just built. Taste Labs can tell you whether that page looks like you. The free Jinn demo token shows you what it feels like when the agent reads your whole verified record first: voice, claims, and design, from one call, with a person still holding the pen on anything that spends. No login, sixty minutes, nothing to install.