Unbrowse skill
A skill, MCP and CLI that give an agent ready-made calls to sites' internal APIs instead of driving a browser; works through the unbrowse.ai cloud service
- 813 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- Signs in to sites with saved logins and can post, send and buy, which is why such actions need your consent
- Site requests come from your IP, while tasks, the index and sessions are handled by an external cloud service
- With a plugin it blocks the agent's built-in web search, page fetching and navigation and routes them through itself
Install
- Works in
Manual install
npx skills add unbrowse-ai/unbrowse --skill unbrowseInstalls the skill. After that you need to sign in: npx unbrowse login opens the OAuth page in the browser and asks you to paste nothing.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Signs in to sites with saved logins and can post, send and buy, which is why such actions need your consent
- Site requests come from your IP, while tasks, the index and sessions are handled by an external cloud service
- With a plugin it blocks the agent's built-in web search, page fetching and navigation and routes them through itself
This is third-party code. Review the repository files before installing.
What it does
Unbrowse connects an agent to the unbrowse.ai cloud service, which turns actions on websites into ready-made calls to their internal APIs, so the agent does not push a page through a browser. The open part of the repository is the client: the `@unbrowse/sdk` TypeScript SDK, the `unbrowse` command line, a skill and plugins. Task execution, the site index and sessions live on the service's servers. A task takes three steps. First `unbrowse discover` looks for a matching capability among your private ones and then in a public registry of sites already compiled into tools. If one matches, `unbrowse run` replays the site's own HTTP requests with no browser and counts a run as successful only when the result is verified. If nothing matches, the agent does the task in a recorded cloud browser through MCP, or you hand over two HAR recordings with `unbrowse learn a.har b.har` and the service compiles them into a one-call capability. Next time the second step fires. There are three ways to connect: the skill `npx skills add unbrowse-ai/unbrowse --skill unbrowse`, MCP with `claude mcp add --transport http unbrowse https://unbrowse.ai/mcp` and OAuth sign-in, or the command line through `npx unbrowse login`. Plugins for Claude Code, Codex, Grok Build, OpenClaw, Hermes and elizaOS make Unbrowse the host's browser.
Who it is for. For developers and analysts whose agents regularly read and fill in sites with no public API and who are slowed down by launching a browser for every step.
Good fit when
- The agent needs data from a site with no public API, and the site is already in the public registry of ready tools
- You want to read a page as Markdown with one command, `unbrowse scrape <url>`, with no browser control
- You recorded two HAR files of the scenario you need and want to turn them into a one-step call
- Requests to the site should come from your IP rather than the service's cloud, which is how `unbrowse run` works by default
Not a fit when
- You need a tool with no sign-up and no cloud service: without signing in, commands exit with code 3, and execution and the index stay on the server
- The site has an official API or MCP server: connecting those directly is simpler than teaching the site through Unbrowse
- You need to capture an app's traffic locally and get an API client as code: Reverse API Engineer in the catalog does that
- The agent has to act on a site without your knowledge: the skill requires the user's consent for posting, sending and purchases
Example request
Use Unbrowse to get the top stories from Hacker News and show titles with linksLimitations
Limits and prices depend on the account plan, and `unbrowse usage` shows the quota. Sign-in is OAuth through an emailed link, and automation and CI use a key in the `UNBROWSE_API_KEY` variable. The MIT license covers the open client, while execution and the index run on the service's server.
The first request to a site may find no ready route, and then the agent gets a `no_capability` status: a ready call appears only after indexing. The skill tells the agent not to promise coverage of every site. The service indexes through a model, and when credits run out the job fails with a `model_unavailable` error.
With a plugin, built-in web search, page fetching and navigation to external addresses are disabled in favor of Unbrowse tools, and `UNBROWSE_ALLOW_BUILTIN_BROWSER=1` brings them back for a session. Use the tool only on sites and data you have permission to work with, and the sites' terms stay on you.
How to disable. Remove the unbrowse server from the MCP list (in Claude Code the `/mcp` command) and disable the plugin through `/plugin`. The command-line token lives in `~/.config/unbrowse/cli.json`: delete the file or sign out on the service page. Saved site logins are kept by the service itself.
FAQ
How is Unbrowse different from driving a browser?
Instead of clicking through a page, the service replays the requests to the site's internal API that the page itself makes. The result is verified: a run counts only with a confirmed answer. The service's cloud browser is needed when there is no ready route yet.
Which IP do site requests come from?
Yours by default: `unbrowse run` sends requests from your machine. The `--from-unbrowse` flag moves the run to the service's cloud.
What do the command-line exit codes mean?
Code 0 means success, 1 an error, 2 a request for input, 3 a needed sign-in, 4 an unverified result. On code 2 you answer with `unbrowse resume <runId> k=v` on the same run.
Where does the service keep site passwords?
In Unbrowse's own password manager. No password is typed into the command line: sign-in is requested with the `credentials request` and `credentials status` commands.
What if a site has no ready tool?
Start indexing with `unbrowse index <url>` and wait for the job status. If you need the answer right away, read the page as Markdown with `unbrowse scrape <url>`.
Discussion
No questions or reviews yet.
Related
SkillsBrowserbase's official skill set that gives Claude Code web access: browser automation, search, page fetching and UI testing
A web research skill: SearXNG searches, Camofox reads pages and CloakBrowser opens sites that check their visitors
Hyperbrowser MCP
Hyperbrowser MCP Server
The official MCP server for Hyperbrowser cloud browsers: page scraping and crawling, structured data, Bing search and three browser agents
A skill and CLI from Browser Use: the agent drives your real Chrome over the DevTools Protocol, logged in as you, and writes any missing helper itself
CLI, skill and Python library for browser control: the agent clicks, fills forms and reads pages over CDP
Web Access
web-access
A web access skill that picks between search, page fetch and driving your own browser over CDP
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Unbrowse (this entry) | yes | 813 | ||
| Reverse API Engineer | no | 2,062 | a local CLI that records a site's traffic into HAR and writes an API client in code in a language you choose | |
| Firecrawl MCP Server | yes | API key | 7,723 | an official scraping MCP server: search, pages to Markdown, site crawling and page actions |
README in short
The repository has a short description of the service, install commands for the skill and MCP, the three steps of operation, a command list with exit codes, an SDK call example, a section on agents and plugins and instructions for building the client with bun. Details are moved into the docs folder: installation, command line, MCP, SDK, pricing, troubleshooting and a paper on internal APIs.
README badge
Original SKILL.md
--- name: unbrowse description: Search and call websites through Unbrowse's hosted API or remote MCP, reuse indexed site tools, read pages, and learn missing routes in its cloud browser. Use for structured website tasks, and authenticated site access with Unbrowse. --- # Unbrowse Use the hosted service at `https://unbrowse.ai`. Execution, indexed routes and website sessions stay server-side. This skill supplies operating guidance; signing in is OAuth, below. ## Onboarding: sign in with OAuth Sign-in is OAuth in the person's browser: an emailed sign-in link, then **Allow**. No password, and nothing to paste. Never ask the person for a token or key in chat. Check first, then sign in only what is missing: 1. MCP: if `unbrowse_discover` is in this session's tool list, the MCP is signed in. 2. CLI: `npx unbrowse whoami`. Exit 0 means signed in. Exit 3 means not signed in. Sign in the MCP (agent hosts): ```sh claude mcp add --transport http unbrowse https://unbrowse.ai/mcp ```