Unbrowse is live · free plan, paid from $10/10K verified calls · last verified 24 Oct 2026 · viability healthy
Unbrowse
Unbrowse turns websites into typed, replayable JSON routes your AI agent calls through a remote MCP server instead of scraping pages.
Worth testing if your agent keeps hitting the same dozen sites: compile once, then call a typed route for 0.1s instead of re-parsing a DOM. The beta's 500 free verified calls a month is enough to find out whether your targets expose anything. The coverage caveat is Unbrowse's own — no official API on 85% of 425 sites — so bench it against your real list before you rely on it.
Verdict generatedVerified 6d ago · liveness 75/100 · cite: rightaichoice.com/tools/unbrowse
- Developers whose agents keep calling the same sites and want a typed JSON route instead of DOM parsing
- Claude, ChatGPT, Cursor or Claude Code users who want web actions from a one-link remote MCP server
- Teams cutting per-run token and browser cost on high-volume repeat tasks
- Builders who need delegated account actions bounded by a vault and per-connection gate
- Non-technical users who want a point-and-click scraping tool
- Workflows needing pixel-identical visual rendering of a page
- Teams that require guaranteed coverage on every site — Unbrowse reports named failures in its own benchmarks
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip Unbrowse if each agent run targets a different, unfamiliar site — with no reusable route, every task falls through to the cloud browser and you lose the replay economics that justify it.
The free 500 verified calls a month is a beta number; the published post-beta rate is $10 per 10,000 verified calls, so plan the meter before you put it in production.
At 500 free verified calls a month during beta, Unbrowse costs nothing to trial with an agent that runs a few dozen site actions daily. After beta, $10 per 10,000 verified calls puts repeat site reads well under the per-token cost of re-parsing pages, and under managed browser-automation minutes priced by the hour. Enterprise and private execution planes are custom-quoted on the same meter.
In short
Unbrowse — Unbrowse turns websites into typed, replayable JSON routes your AI agent calls through a remote MCP server instead of scraping pages. Best for Developers whose agents keep calling the same sites and want a typed JSON route instead of DOM parsing, Claude, ChatGPT, Cursor or Claude Code users who want web actions from a one-link remote MCP server, Teams cutting per-run token and browser cost on high-volume repeat tasks. Free to start; paid plans from $10/10K verified calls.
What is Unbrowse?
Unbrowse — Unbrowse turns websites into typed, replayable JSON routes your AI agent calls through a remote MCP server instead of scraping pages. Best for Developers whose agents keep calling the same sites and want a typed JSON route instead of DOM parsing, Claude, ChatGPT, Cursor or Claude Code users who want web actions from a one-link remote MCP server, Teams cutting per-run token and browser cost on high-volume repeat tasks. Free to start; paid plans from $10/10K verified calls.
Unbrowse is an action layer for AI agents, shipped as one remote MCP endpoint at unbrowse.ai/mcp. Paste the link into Claude, ChatGPT, Cursor, Claude Code or any MCP-capable client, sign in with OAuth, and the assistant can search a site, read a page, or perform an action there. Execution happens in Unbrowse's cloud, so nothing installs locally. The engine is what separates it from a headless-browser wrapper. Instead of re-rendering a page on every request, Unbrowse watches the first-party requests a site makes and compiles those into a versioned harness with typed input and output schemas. Replayed routes answer over ordinary HTTP — the homepage shows hn.algolia.com returning in 0.09s and a docs.rs search route in 0.14s — with a cloud browser kept as the fallback when no route exists yet. Compiled tools land in a public registry with health hints, so the next run is a lookup rather than a scrape. An October 2026 Unbrowse benchmark of 425 sites found no official API covering the task on 85% of them, which is the gap it targets. Developers get the same runtime through the @unbrowse/sdk TypeScript package, a Python SDK, canonical REST under /api/v1 with OpenAPI 3.1, Agent Skills exposing unbrowse_skill_* tools, and per-site MCP servers. A password vault autofills saved logins and 2FA into a site's own form without handing the secret to the model, and a per-connection gate can allow, deny or ask before an account is touched. Metering is tied to verified dispatch: one billable call equals one successful dispatch through a harness operation, and policy denials, failed grants and destination blocks don't count. That suits teams whose agents hit the same sites repeatedly and would rather call a typed endpoint than burn tokens parsing a DOM. Universal coverage is the trade — stable targets pay off, wandering ones don't.
Key features
- Remote MCP server at /mcp over Streamable HTTP
- OAuth sign-in to connect an assistant, no API key
- Search a site, read a page, or act on it as separate capabilities
- Shadow API discovery from a site's own first-party request traffic
- Compiled routes replayed over plain HTTP (first_party_http)
- Versioned harness with typed input and output schemas
- TypeScript SDK (@unbrowse/sdk) with run, resume and idempotency keys
- Python SDK for the same runtime
- Canonical REST API under /api/v1 with OpenAPI 3.1
- Agent Skills exposing unbrowse_skill_* tools with slot schemas
- Per-site MCP server plus OpenAPI 3.1 document per compiled site
- Cloud browser fallback with snapshots when no route exists
- Password vault that autofills logins and 2FA without exposing secrets to the model
- Per-connection gate to allow, deny or ask before account access
- Public registry of compiled tools with health hints and p50/p95
Behind the verdict
The pitch is narrow and honest: stop having your agent scrape the same page for the hundredth time. What Unbrowse builds instead is a compiled, versioned harness per site, so a request becomes a typed JSON call dispatched over first-party HTTP. In practice that means latency in the tenths of a second and a far smaller token bill than shipping HTML into context. Pick this when your workflow has a stable target set — a handful of sites you query daily, a docs search, an internal portal, a flight or product lookup. That's where the replay model earns its keep. The public registry and health hints matter here too, since a route someone already compiled is a lookup rather than a fresh index. Pass when your agents roam. If every task lands on a different site, you're on the cloud browser fallback each time and lose the entire advantage. Teams needing pixel-perfect rendering of a page are also in the wrong tool. The closest alternative is a conventional browser-automation stack like Playwright, often wrapped in an MCP server. Playwright covers more ground and doesn't ask you to trust anyone's compiled routes. Unbrowse trades that breadth for repeatability and speed, which is the right trade only when the repetition is real. Two things I'd watch. First, freshness: a compiled route drifts when the site changes, so the health hints and browser fallback aren't decoration — they're the safety net. Second, the account access path. The vault and per-connection gate are the parts to scrutinize hardest if agents will be acting inside logged-in accounts, not just reading. We'd start with read and search routes on non-critical sites, confirm the verified-call meter behaves as documented, and only then wire in account actions. Beta pricing makes that experimentation cheap; the $10
Limitations
- Unbrowse is a routing and compilation layer, not a model provider, so output quality depends on the agent driving it and on how much of a site's first-party API is discoverable.
- Its own published benchmark of unbrowse@6.17.2 reports 70% coverage across 37 probes with named failure rows, so some sites never compile.
- When a route is missing the patchright browser fallback re-introduces conventional browser cost and fragility, and a page that hits a bot wall returns blocked for 10 minutes with a retryAfter.
- Scrapes fail with scrape_timeout at the deadline (default 45s, 5-120s range).
- The DOM-to-route token saving is stated qualitatively, not as a measured ratio.
- It is in beta: 500 verified calls a month are free and the published post-beta rate is $10 per 10,000 verified calls.
as of 2026-10-19
Researching Unbrowse? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Is Unbrowse still active?
Unbrowse is live · free plan, paid from $10/10K verified calls · last verified 24 Oct 2026 · viability healthy.
Viability score
How well maintained and how widely used is Unbrowse? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: October 2026
How we score →What's new in Unbrowse
Checked 10 days agoAcross the latest 4 updates: 4 news mentions.
The published API vs. the API a site's own pages call
Unbrowse argues agents should target the internal API websites call from their own pages, since published APIs can be switched off at any time.
Whitepaper: 'Pay for the Delta' proof of indexing
A whitepaper proposes per-use payments to index maintainers based on the improvement their route made, explicitly with no stake, no token and no land grab.
Third parties can learn a site's routes without seeing queries
A technical post outlines delegating route learning to other machines while keeping the requester's query private, alongside a plan to default to the user's already-signed-in browser over cloud browsers.
Benchmark: no official API on 85% of 425 sites checked
Unbrowse checked 425 sites agents already work with and reports that on 85% no official API covers the task; the post includes its method and stated limits.
What do users say about Unbrowse?
What people actually say about Unbrowse — is it worth it?
We ran a structured research pass across product reviews, community discussions, and post-purchase forum threads to surface the patterns vendors won't publish themselves. Below: the recurring strengths, the hidden costs people mention most, and the cohort that consistently regrets adopting this tool.
ResearchedDiscussed by users across 3 sources (Hacker News, YouTube, GitHub) · researched Sep 9, 2026.
Weighted by the 39 posts each of 3 sources contributed.
- +Massive speed improvements—3.6x faster than Playwright in tests.
- +Cost reduction up to 80% by bypassing headless browsers.
- +Innovative shadow API discovery that learns routes automatically.
- +80% of browser traffic can be avoided on covered sites.
- +Proactive maintainers who close most GitHub issues within days.
- −Security scare with an npm plugin that exfiltrated credentials.
- −No support for multiple account profiles or switching between them.
- −24-hour route cache lacks per-call freshness checks, risking stale data.
- −Browser fallback quality is uneven on authenticated or dynamic sites.
- −Free tier unclear—users report not seeing a free option.
- • Users may incur charges for heavy browser fallback usage or for retries that double-settle payments.
- • Route caching and reliability extend traffic, which could escalate credit consumption on high-traffic sites.
How we verified this
Every listed fact about Unbrowse— pricing, features, links, status — is re-read from the vendor's own pages by our verification pipeline and each change is logged with a date; the last pass ran on 24 Oct 2026. Pages mix model-assisted synthesis with human review; see how we verify.
Verification history
We have re-verified Unbrowse 9 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — changed our verdict, who it suits, features, description, who should skip it, pricing tiers, summary
- — changed limitations, summary, our verdict, who should skip it, integrations, setup time, use cases, hidden costs, features, switching out, who it suits, description, switching in, pricing fit
- — changed SSL certificate
- — changed who should skip it, who it suits, pricing tiers, summary, our verdict, features, description
- — changed pricing page
- — changed SSL certificate, our verdict, integrations, summary, pricing tiers, who should skip it, features, who it suits, description
Showing the 6 most recent of 9 verification passes.
Free to cite with attribution — this page re-verifies continuously.
How much does Unbrowse cost in 2026?
Unbrowse: Free to start; paid plans from $10/10K verified calls. Pricing last verified 24 Oct 2026.
Plans compared
Every published Unbrowse tier, its price as the vendor publishes it, who it fits, and what it adds over the tier below. The 12-month cost projection follows.
| Plan | Price | Billing | Who it's for |
|---|---|---|---|
| Beta | Free | — | — |
| After beta (usage) | $10/10K verified callsVendor: “$10 per 10,000 verified calls” | usage | Teams moving an agent with a stable target list into production once the beta ends.Adds: Replaces the free allowance with metered usage at $10 per 10,000 verified dispatches on the same meter. |
| Enterprise / customer-private execution plane | Custom | quote | Organizations that need captures, vault material and inference to stay inside their own contractual boundary.Adds: Same per-call meter but different custody — private execution plane instead of the shared cloud environment. |
12-month cost
Project the real annual outlay, including the implied monthly cost when only an annual tier is published.
Vendor list price only. Add-on usage, seat overages, and contract minimums are surfaced under Hidden costs & gotchas.
Where the pricing makes sense
The company stage and team size where Unbrowse's pricing actually pencils out — and where peers do it cheaper.
At 500 free verified calls a month during beta, Unbrowse costs nothing to trial with an agent that runs a few dozen site actions daily. After beta, $10 per 10,000 verified calls puts repeat site reads well under the per-token cost of re-parsing pages, and under managed browser-automation minutes priced by the hour. Enterprise and private execution planes are custom-quoted on the same meter.
Has Unbrowse's pricing changed?
No plan price for Unbrowse has moved since we began logging in Aug 2026; the pricing page is re-read on every verification pass.
Who should pick Unbrowse?
Pick Unbrowse if you are Developers whose agents keep calling the same sites and want a typed JSON route instead of DOM parsing, Claude, ChatGPT, Cursor or Claude Code users who want web actions from a one-link remote MCP server, Teams cutting per-run token and browser cost on high-volume repeat tasks, Builders who need delegated account actions bounded by a vault and per-connection gate.
Look elsewhere if you are Non-technical users who want a point-and-click scraping tool, Workflows needing pixel-identical visual rendering of a page, Teams that require guaranteed coverage on every site — Unbrowse reports named failures in its own benchmarks, One-off tasks across a different site each time, where only the cloud browser applies.
Skip it if Skip Unbrowse if each agent run targets a different, unfamiliar site — with no reusable route, every task falls through to the cloud browser and you lose the replay economics that justify it.
GeneratedReal-world workflow fit
Concrete scenarios for the personas Unbrowse actually fits — and what changes day-one when you adopt it.
You paste the remote MCP link once, sign in with OAuth, and ask your assistant for a repo's README. It calls a compiled GitHub route rather than scraping the rendered page.
Outcome: Reads come back as typed JSON in a fraction of the time of a DOM scrape, and the route stays reusable for every later run against that site.
You call unbrowse_run with capability skyscanner.flights and inputs SIN to ICN via @unbrowse/sdk, passing an idempotency key and unattended interaction mode.
Outcome: A durable run returns a verified result; if the site needs an extra field, resume answers it at the returned state revision instead of restarting the task.
You save a login through the password manager, and the agent calls unbrowse_browse_act with vault autofill on a site's own form while the per-connection gate asks you once.
Outcome: The action runs under your authenticated session while the model only ever sees masked hints, not the credential.
Use cases
- Pull live flight prices from Skyscanner by calling its own first-party search API instead of rendering the page.
- Read a GitHub repo's README through a compiled route rather than scraping the rendered DOM.
- Fetch top stories from Hacker News via top_stories over plain HTTP for a daily digest agent.
- Compile a restaurant site such as eatigo.com/sg into a learned get_search capability your agent replays.
- Serve every compiled site to your agent as a per-site MCP server and wire it into Cursor or Claude.
- Index a whole site ahead of need so background jobs pre-build read-only tools before your agent asks.
- Have the vault type a saved login into a site's own form so an agent can act behind an authenticated session.
- Review recorded cloud-browser sessions as a text timeline to find where an agent run dropped off.
Setup time & first value
How long it actually takes to get something useful out of Unbrowse — broken out by persona, not the marketing-page minute.
For a developer already in Claude, ChatGPT or Cursor, first value is a pasted MCP link plus an OAuth sign-in — a few minutes. Wiring the TypeScript or Python SDK against /api/v1 with an org key is an afternoon including the OpenAPI spec read. Compiling a usable route set for a specific target list depends on the sites; expect a day of probing before you know your coverage.
Switching to or from Unbrowse
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From Playwright: replace per-page scripting with unbrowse_run against a compiled capability for your stable targets, keeping Playwright for the long tail.
- →From Firecrawl: use unbrowse_scrape for one-page markdown plus metadata.status and timing, and move repeat reads to compiled routes.
- →From a hand-rolled internal API client: feed captured HAR files to unbrowse_learn and get a one-call learned.* capability with a typed schema.
- ↗To Playwright: export your target list and re-implement the routes as browser scripts where coverage gaps remain.
- ↗To Browserbase or a managed browser: keep your agent code and swap the execution layer for a hosted session API.
- ↗To Firecrawl: for page-to-markdown extraction only, point your fetch calls at its scrape endpoint.
Integrations
Resources & guides
- Documentationunbrowse.ai
Docs · Unbrowse
Full product docs from unbrowse.ai
- Quickstartunbrowse.ai
Quickstart · Unbrowse
Get up and running fast from unbrowse.ai
- Resourceunbrowse.ai
Openapi · Unbrowse
Helpful link from unbrowse.ai
- Resourceunbrowse.ai
Llms · Unbrowse
Helpful link from unbrowse.ai
- Resourceunbrowse.ai
Pricing · Unbrowse
Helpful link from unbrowse.ai
Tutorials & Learning
YouTube returned 6 videos for “Unbrowse”, and we withheld 6: 6 could not be judged, because “Unbrowse” is a single word that other videos use for other things. We are showing none, because we could not prove any of them are about Unbrowse.
Official links
What are the best Unbrowse alternatives?
The closest alternatives to Unbrowse we list are Browserbase, Firecrawl and Safari Mcp. Ranked by direct product-type match, not category overlap.
| Alternative | Pricing | Starts at | Last verified |
|---|---|---|---|
| Browserbase | Freemium | Pricing not published | — |
| Firecrawl | Freemium | Pricing not published | — |
| Safari Mcp | Contact Sales | Custom pricing | — |
Alternatives to Unbrowse
View all
Browserbase
Give AI agents cloud browsers to run the web like an API — logins, CAPTCHAs and forms handled.
Firecrawl
Firecrawl is a web data API that scrapes, searches, and interacts with any page or PDF to feed LLM-ready Markdown and JSON into AI agents.
Safari Mcp
Mac-only MCP server that drives your installed Safari via AppleScript so AI agents can browse, click, and extract pages
Tools that pair well with Unbrowse
Common stack mates teams adopt alongside Unbrowse, with the specific reason each pairing earns its keep.
Browserbase
Give AI agents cloud browsers to run the web like an API — logins, CAPTCHAs and forms handled.
Firecrawl
Firecrawl is a web data API that scrapes, searches, and interacts with any page or PDF to feed LLM-ready Markdown and JSON into AI agents.
Safari Mcp
Mac-only MCP server that drives your installed Safari via AppleScript so AI agents can browse, click, and extract pages
Unbrowse vs its closest rival
Unbrowse's closest head-to-head is with Temporal Ai — If your AI agent needs to survive crashes and orchestrate multi-step workflows with automatic retries and rollbacks, Temporal is the clear choice — OpenAI and Lovable trust it. If your goal is to extract web data at speed and low cost by calling internal APIs instead of running browsers, Unbrowse delivers up to 70% coverage with 96% avoidance rate. Most technical buyers will need both: Temporal for durable orchestration, Unbrowse for efficient data ingestion. Verdict generated
Head-to-head comparisons
Unbrowse vs Temporal Ai
If your AI agent needs to survive crashes and orchestrate multi-step workflows with automatic retries and rollbacks, Temporal is the clear choice — OpenAI and Lovable trust it. If your goal is to extract web data at speed and low cost by calling internal APIs instead of running browsers, Unbrowse delivers up to 70% coverage with 96% avoidance rate. Most technical buyers will need both: Temporal for durable orchestration, Unbrowse for efficient data ingestion.
GeneratedUnbrowse vs Spider Cloud
If your goal is to build autonomous agents that call websites' internal APIs directly for speed and cost savings, Unbrowse is the innovative choice—its shadow API discovery and shared route graph can drastically reduce browser overhead. However, if you need to crawl at massive scale (100K+ URLs per request) with robust anti-blocking and structured data extraction for RAG pipelines, Spider Cloud is more mature and feature-rich, with a Rust engine, Silk AI extraction, and a new flat-rate plan. Choose Unbrowse for bleeding-edge API-level extraction; choose Spider Cloud for reliable, high-volume web data.
GeneratedUnbrowse vs Presto Voice
If you run a QSR chain and want to boost drive-thru revenue with minimal staff overhead, Presto Voice is the proven turnkey choice. If you're building AI agents that need to interact with web services efficiently, Unbrowse's API-first approach slashes costs versus browser automation. They serve completely different buyers — choose based on your domain.
GeneratedFrequently asked questions
Used Unbrowse? Help shape our editorial sentiment research.