AginxBrowser
AginxBrowser
El navegador para agentes de IA. Ve la web en vivo. Interactúa con ella.
Un navegador construido para agentes desde la primera línea de código, no un navegador humano acoplado a la automatización. Ve el mundo, léelo, búscalo y actúa sobre él: un solo binario de Rust con V8 integrado, sin necesidad de Chromium.
Los humanos tienen Chrome. Los agentes tienen AginxBrowser.
Un solo binario, cero dependencias, servicio instantáneo. API HTTP + MCP nativo: los agentes se conectan y listo.
Por qué los agentes necesitan su propio navegador
La «automatización de navegadores» existente se creó para humanos o para extracción de una sola pasada, no para agentes:
AginxBrowser | Puppeteer/Playwright | Firecrawl | Browser-use | |
Diseñado para | Agentes primero | Depuración humana | Servicio de scraping | Envoltorio de LLM |
Dependencias | Un solo binario, sin Chromium | Chromium ~500MB | Docker ~1GB | Chromium |
Ve (capturas de pantalla) | ✅ motor de renderizado diting integrado | Necesita Chromium | ❌ | Necesita Chromium |
Lee | markdown + js_extract | DIY | markdown | DIY |
Busca (búsqueda) | ✅ metabúsqueda de 5 motores | ❌ | ❌ | ❌ |
Actúa | interacción de sesión indexada | API de DevTools | ❌ | Impulsado por LLM |
Protocolo | HTTP + MCP nativo | API de Node | HTTP | Python |
Huellas TLS | ✅ Chrome/Firefox/Safari | Requiere plugin | ❌ | ❌ |
Resolución de CAPTCHA | ✅ automática | DIY | ❌ | ❌ |
Sesiones interactivas | ✅ persistentes | ✅ | ❌ | ✅ |
Un agente necesita cinco cosas de un navegador: ver, leer, buscar, actuar, desplegar. Un solo binario lo cubre todo: compatible con systemd, MCP nativo para Claude/Cursor, cero dependencias.
Ventaja principal: sin Chromium. AginxBrowser integra un motor de navegador completo (V8 + pila HTTP de Rust + nuestro propio motor de renderizado diting de CSS/diseño/pintura, con el linaje de Blitz/Stylo/Taffy como implementación de referencia). Sin Puppeteer, sin Chrome, sin Docker. Un solo binario de Rust bajo systemd es tu infraestructura de navegación para agentes.
Related MCP server: Web Bridge
Tres cosas que los renderizadores sin estado no pueden hacer
La mayoría de los nuevos «navegadores de agentes» son renderizadores de una sola pasada, sin estado y sin huellas digitales; sirven para páginas públicas, pero mueren al instante frente a Cloudflare o flujos de inicio de sesión. AginxBrowser va en la dirección contraria:
🔐 Huellas TLS reales — el modo sigiloso replica los protocolos de enlace TLS completos de Chrome145 / Firefox133 / Safari / Edge mediante BoringSSL (no solo una cadena UA), conmutables por petición; los desafíos de Cloudflare Turnstile esperan automáticamente a
cf_clearance. Los motores sin huellas reciben 403 — nosotros pasamos.🤝 Sesiones interactivas con estado — sesiones persistentes (keep-alive de 8 minutos en inactividad), estado de inicio de sesión inyectable y exportable (
session_create(cookies=...)↔session_cookies), que sobreviven a la paginación y a flujos de varios pasos. Los motores de una sola pasada desechan el estado.🔌 MCP nativo — 14 herramientas de primera clase (no un shim de CDP). Claude Code / Cursor / Claude Desktop se conectan en una línea. Protocolo dual HTTP + MCP.
Punto de referencia: Kitesurf de Cloudflare no incluye explícitamente ni negociación real de huellas TLS ni sesiones de autenticación persistentes; el territorio antibot y de inicio de sesión es exactamente donde juega AginxBrowser.
Código abierto Apache-2.0, un solo binario: autohospeda hoy, sin dependencia de la nube.
Capacidades
Renderizado por niveles: páginas estáticas mediante HTTP simple (~100ms); V8 se activa solo cuando se necesita renderizado JS (~1-2s): 10 veces más rápido en ~80% de las páginas.
Metabúsqueda de múltiples motores: web general (Baidu / Bing / Sogou / WeChat / Google), noticias (Bing News), código (Stack Overflow, GitHub), paquetes (npm, PyPI), académico (arXiv), modelos de IA (Hugging Face) — consultados en paralelo, fusionados y deduplicados. Los operadores pueden conectar un índice privado de Meilisearch al mismo
/search. Buscar → leer en un solo paso.Búsqueda de imágenes:
categories=imagesconsulta los índices de imágenes de Baidu/Bing y devuelve enlaces binarios directosimage_url(descargables directamente a jpg/png) además de la procedenciasource_url.Sesiones interactivas: sesiones de navegador persistentes con interacción indexada (
state/click/input/scroll/eval): los agentes navegan como lo hacen los humanos.Resolución automática de CAPTCHA: detección de tipos con integración opcional de 2captcha: la búsqueda nunca se detiene en páginas de verificación.
Extracción de datos JS:
js_extractextraewindow.__INITIAL_STATE__y otros datos estructurados de las SPA.Renderizado de capturas de pantalla: el endpoint
/screenshot(opt-in--features screenshot) pinta el DOM renderizado por JS con nuestro propio motor de renderizado diting — solo CPU, sin Chromium — a PNG. Entrada visual para agentes.Espera automática de Cloudflare: detecta las páginas de desafío «Just a moment...» y espera hasta obtener
cf_clearance.Suplantación de huellas TLS: el modo sigiloso suplanta a Chrome145/Firefox133/Safari/Edge, conmutable por petición.
Servidor MCP: el modo
--mcpexpone 14 herramientas (fetch/eval/click/search/download + 9 herramientas de sesión): Claude Code / Claude Desktop / Cursor las llaman directamente.Compatible con Firecrawl: el endpoint
/v1/scrape— los clientes actuales de Firecrawl migran cambiando la URL base.Protección contra rebinding de DNS: guardia SSRF integrada + validación de IP posterior a la resolución.
Para qué sirve
No son demostraciones: trabajos reales que los navegadores de agentes hacen hoy en día:
Abrirse paso por consolas de administración — AWS / App Store Connect / Google Play, docenas de capas de menú por tarea. Deja que el agente haga clic; solo vuelve cuando se necesita autorización.
Acciones masivas tras el inicio de sesión — llenar carritos, escarbar en el historial de pedidos, revisar páginas que solo se renderizan con la sesión iniciada. Inyecta cookies, opera, exporta para reutilizar.
Atravesar muros antibot — protección de Cloudflare, desafíos Turnstile, comprobaciones de huellas TLS. El modo sigiloso avanza en lugar de retroceder ante un 403.
El internet chino — metabúsqueda de Baidu / Sogou / WeChat en 5 motores, renderizado correcto de páginas chinas. No solo web en inglés.
Scripting sobre la marcha — el agente lee la página, escribe JS, lo evalúa: tablas comparativas resaltadas, contenido reajustado, filtros de productos sobre parámetros ocultos. GreaseMonkey con esteroides.
Visión multimodal — capturas de pantalla como entrada visual para flujos de mirar y decidir: elegir asientos, reconocer diseños, verificar el renderizado.
Inicio rápido
Prueba primero la instancia alojada: https://browser.aginx.net/
Instalación completa con un solo comando (superficie de activación SKILL.md + herramientas MCP + verificación):
# Download -> inspect the contents -> run only after review (never blind-run network scripts)
curl -fsSL https://raw.githubusercontent.com/yinnho/aginxbrowser/main/skill.sh -o skill.sh
less skill.sh
bash skill.shRegistrar solo MCP:
claude mcp add aginxbrowser --transport http https://browser.aginx.net/mcpInstala la superficie de activación de la skill mediante skills.sh:
npx skills add yinnho/aginxbrowserAutohospedaje:
# Prebuilt binary (one-line installer: platform detect + sha256 + self-check)
curl -fsSL https://raw.githubusercontent.com/yinnho/aginxbrowser/main/install.sh -o install.sh
less install.sh && bash install.sh
aginxbrowser doctor # features + fonts + egress self-check
# Or build from source (--features stealth,screenshot or you lose both)
cargo build --release --features stealth,screenshot
# Start the service
./target/release/aginxbrowser
# → Listening on 0.0.0.0:8089
# Verify
curl http://127.0.0.1:8089/health
# → {"status":"ok","engine":"diting"}
# Fetch a page
curl -sS -X POST http://127.0.0.1:8089/fetch \
-H "Content-Type: application/json" \
-d '{"url":"https://example.com"}'
# Search
curl -sS -X POST http://127.0.0.1:8089/search \
-H "Content-Type: application/json" \
-d '{"q":"macbook price","max_results":5}'
# Create an interactive session
curl -sS -X POST http://127.0.0.1:8089/session/create \
-H "Content-Type: application/json" \
-d '{"url":"https://example.com"}'
# → {"session_id":"s_1","url":"https://example.com/"}
# MCP mode (for AI agents)
./target/release/aginxbrowser --mcpEstructura del proyecto
aginxbrowser/
├── Cargo.toml
├── build.rs # V8 snapshot generation
├── js/
│ └── bootstrap.js # V8 bootstrap script
├── README.md
├── docs/
│ └── API.md # Full API reference (HTTP + MCP)
└── src/
├── main.rs # HTTP service entry & routing
├── server.rs # Business layer (fetch/click/eval/search)
├── session.rs # Interactive browser sessions
├── captcha.rs # CAPTCHA detection & auto-solve
├── render.rs # Tiered rendering (HTTP direct → diting browser engine)
├── mcp.rs # MCP server (14 tools)
├── firecrawl_compat.rs # Firecrawl-compatible /v1/scrape endpoint
├── browser.rs # Top-level API: Browser, BrowserBuilder
├── page.rs # Top-level API: Page, Element
├── config.rs # BrowserConfig
├── cookie.rs # CookieStore
├── error.rs # Error types
├── search/ # Native search engines
│ ├── mod.rs # SearchEngine trait, Registry, merge/dedupe, progressive backoff
│ ├── baidu.rs # Baidu (JSON API, wreq stealth)
│ ├── bing.rs # Bing (HTML parsing, plain reqwest)
│ ├── sogou.rs # Sogou web (HTML parsing, plain reqwest)
│ ├── sogou_wechat.rs # Sogou WeChat (HTML parsing + /link resolution)
│ ├── google.rs # Google (HTML parsing, wreq stealth + proxy)
│ ├── stackexchange.rs # Stack Overflow (SE API v2.3, code category)
│ ├── github_repos.rs # GitHub repos (api.github.com, code category)
│ ├── arxiv.rs # arXiv (Atom API, academic category)
│ ├── bing_news.rs # Bing News RSS (news category; proxy-first)
│ ├── huggingface.rs # HF Hub models/datasets/spaces (ai category)
│ ├── npm.rs # npm packages (npms.io API, packages category)
│ ├── pypi.rs # PyPI name resolution (JSON API, packages)
│ └── meilisearch.rs # Private-index adapter (env-configured)
│
├── diting_dom/ # HTML parsing, DOM tree, CSS selectors
├── diting_net/ # HTTP client, cookies, encoding, proxies
├── diting_js/ # V8 runtime, JS ops, module loading
└── diting_browser/ # Page navigation, lifecycle, browser contextCompilación
# Standard build (no stealth; TLS fingerprint features inactive)
cargo build --release
# With stealth (requires go + cmake + C++ toolchain; enables TLS fingerprint spoofing)
cargo build --release --features stealth
# With screenshot rendering (enables /screenshot; adds the rendering stack, +30-40MB)
cargo build --release --features screenshot
# Full featured (recommended for production)
cargo build --release --features stealth,screenshotRequisitos: Rust 1.78+; la biblioteca estática de V8 se descarga automáticamente en la primera compilación. La funcionalidad de sigilo necesita además go, cmake y un compilador de C++. La funcionalidad de capturas de pantalla incluye un subconjunto de fuentes CJK integrado (GB2312 + símbolos comunes): no se requieren fuentes del sistema para un renderizado correcto del chino.
Variables de entorno en tiempo de ejecución
Variable | Por defecto | Descripción |
|
| Dirección de escucha |
| habilitado |
|
| Linux Chrome145 | User-Agent suplantado |
|
| Cabecera Accept-Language |
| ninguno | Proxy de respaldo opcional. Los motores de fuentes bloqueadas (Google, Bing News, Hugging Face) se conectan directamente primero y solo recurren a este proxy cuando falla el intento directo — los despliegues en el extranjero no necesitan ningún proxy; |
|
| TTL de caché de |
| ninguna | Clave de API de 2captcha; habilita la resolución automática de CAPTCHA |
|
| Proveedor de resolución de CAPTCHA |
| ninguna | URL base de Meilisearch; si se define, habilita el motor de índice privado |
| ninguno | uid del índice de Meilisearch a consultar |
| ninguna | Clave Bearer opcional para la instancia de Meilisearch |
Documentación de la API
Referencia completa de la API → docs/API.md
Notas de auditoría de seguridad → docs/skills-sh-audit.md — por qué skills.sh muestra "Critical Risk" y a qué función real del producto corresponde cada advertencia
Cubre:
Todos los endpoints HTTP (
/fetch,/click,/eval,/search,/v1/scrape, 8 endpoints de sesión)Las 14 herramientas del servidor MCP y sus parámetros
Configuración de los clientes Claude Code / Claude Desktop / Cursor
Variables de entorno, códigos de error, ejemplos de scraping por sitio
Integración con otros sistemas
AginxBrowser es infraestructura puramente adyacente — como un navegador real, se ejecuta como un servicio independiente al que cualquier cosa puede llamar, sin incrustar código del host ni contaminar la configuración del host. Despliega una instancia por máquina (bajo systemd) y todas las aplicaciones que necesiten capacidad de "render + scrape" la comparten.
Integración: lee la variable de entorno AGINXBROWSER_URL=http://127.0.0.1:8089. Sin definir → comportamiento sin cambios; definida → los sitios de riesgo controlado se enrutan automáticamente a través de AginxBrowser para el renderizado, con una degradación elegante en caso de fallo.
Limitaciones conocidas
Las capturas de pantalla son opcionales:
/screenshotrequierecargo build --release --features screenshot(añade el stack de renderizado, +30-40MB). El motor de renderizado por defecto es diting (nuestro propio stack de CSS+layout+paint); pasaengine: "blitz"para volver a optar por el pipeline de referencia de Blitz. El CSS de sitios complejos es aproximado en ambos (no es perfecto píxel a píxel como Chromium)Coordenadas de elementos admitidas (a nivel de bloque):
/screenshotconselectordevuelve las coordenadas de página del elemento (selector_rects, px CSS);selectorsolo recorta directamente ese elemento. Los elementos puramente en línea (<a>text</a>) no tienen una caja independiente — elige un ancestro de bloqueLa interacción con JS funciona en general; las páginas con fingerprinting intensivo pueden seguir fallando: la delegación de eventos de React/Vue funciona con normalidad (los atributos de reflexión de URL como
src/hrefse resuelven a URLs absolutas, por lo que Next.js/webpack hidratan la página y los clics disparan los manejadores). Las páginas de autenticación con fingerprinting intensivo (WorkOS/Cloudflare) que sondeannavigator.plugins, el canvas de WebGL, etc. pueden seguir fallando hasta que se complete la cobertura de fingerprinting del modo sigilosoSoporte de proxy: HTTP/HTTPS/SOCKS5 mediante
AGINXBROWSER_PROXYSitios difíciles de riesgo controlado: Baidu Wenku no está soportado; los artículos de Zhihu necesitan un
__zse_ckválido
Licencia
Coherente con el proyecto principal OpenCarrier. Apache-2.0.
Available Tools
40 toolsaccount_deleteAInspect
Delete a named login identity: stored record AND live jar. Cookie values are credentials — delete means gone. Sessions currently running as the account keep their in-process jar handle, but nothing writes back. Returns {deleted: name}, or an error naming the account if it does not exist.
Inputs: name (required).
account_listARead-onlyInspect
List named login identities (the multi-account layer) with metadata only: name, cookie domains, cookie count, updated_at, the last account_verify verdict, and the identity's persona User-Agent (each account is one stable device: its own UA and hardware fingerprint, drawn once and reused). Cookie values are credentials and never leave the server. Use to see which identities exist before session_create {account} picks one.
Inputs: no named parameters.
account_loginAInspect
Open a site's login page AS a named account and close the login loop. Creates the session as the account (private jar, device persona), navigates to url, and reports which generic login gates the page shows — needs: password | sms | qr | slider (QR scan like xiaohongshu, SMS code, password form, slider/captcha) — plus a session_verdict classification and, when a human step is needed, the session_id and a /live handoff so a person can finish it in the live view (the engine detects and describes; it never fills credentials or solves challenges). With predicate and no human gates detected, waits for the automatic bounce, then teaches + stamps the account's verify spec — later account_verify calls re-check it bare. Cookies write back after every action, so a login finished in /live is already persisted. status is "logged_in" (verify stamped), "waiting" (drive session_wait/session_input on session_id, or re-call account_login after the human finishes — the account's shared jar already holds it), or "opened" (no predicate given).
Inputs: url (required), name (required), predicate, use_proxy, +1 more.
account_verifyAInspect
Check whether a named account is still logged in. Teach-once: the first call passes url + predicate (a JS expression truthy on a logged-in page, e.g. !!document.querySelector('.user-nick')); the spec is remembered and later calls can be bare. Runs in a scratch session AS the account (private jar), so the probe doubles as a cookie refresh. Returns {name, logged_in, url, checked_at}.
Inputs: name (required), url, predicate.
cacheAInspect
Query the LOCAL CACHE of every page this server has fetched and every search it has run. Check here BEFORE re-fetching or re-searching — a hit is instant and free while a fresh fetch costs 5-60s. Use query for full-text search (works for Chinese substrings and English words), get to pull a page's full cached content, stats for counts, clear to delete rows.
Inputs: all, get, url, kind, +5 more.
clickAInspect
Click an element on a one-off page: loads url in a fresh browser context (stateless — no cookies unless passed, no shared state with other calls), waits wait_secs after load before clicking, then fires a DOM click on the first CSS-selector match. The click may trigger navigation (link, form submit) — the response url and text_after are read after that navigation lands. Returns clicked:false when the selector matches nothing. For multi-step interaction on a shared page use session_click instead.
Inputs: url (required), selector (required), wait_secs.
downloadAInspect
Download a file over HTTP(S) with streaming to disk (no memory buffering), SHA-256 integrity hash, and optional resume of interrupted transfers. Filename resolution: explicit param → Content-Disposition → URL tail. Use for binaries, archives, datasets, documents - anything where the agent wants the FILE saved, not its text content read.
Inputs: url (required), resume, cookies, filename, +1 more.
evalAInspect
Execute JavaScript on a one-off page: loads url in a fresh browser context, optionally waits wait_secs for the page to settle, evaluates script (async/Promise supported) and returns {url, result}. Script-driven navigation (location.href, form submit) is drained and reflected in the returned url. Stateless — no cookies or page state shared with other calls; when the script needs prior page state or a login, use session_eval.
Inputs: url (required), script (required), wait_secs.
fetchARead-onlyInspect
Fetch a webpage and return clean markdown/html/text. Use whenever the agent needs to READ any web page - blogs, docs, articles, JS-rendered SPAs, Cloudflare-protected sites. Static pages are served over plain HTTP (~100ms tier:"http"); pages that need JS get the full browser (tier:"browser"). render_tier selects auto (default) / http (pure HTTP, refuses the upgrade) / browser (always the JS browser).
Inputs: url (required), format, sanitize, selector, +8 more.
flow_runAInspect
Run a flow — a recorded, editable JSON browser-session script — deterministically, with zero model tokens. Steps are {op, args, expect?, save?}: ops cover navigate/click/click_xy/input/scroll/eval/wait/screenshot/state/cookies; {{var}} placeholders in args are filled from vars; expect asserts (url_contains | selector | text_contains | eval_truthy) abort with evidence on failure; save collects a step's output into the receipt. Source the flow inline via "flow", or by "name" from the server's workflow//flow.json (unknown name → error lists installed workflows). Pass session_id to reuse a live session (e.g. from import_curl) so login state and flows compose. The receipt carries status ok/failed, saved outputs, the session_id (kept alive), and on failure the failing step, reason and a diagnostic screenshot — fix the flow or take the session over from there.
Inputs: flow, name, vars, max_steps, +1 more.
import_curlAInspect
Import login state from a real browser in one paste. The human logs into a site in their own Chrome (solving the CAPTCHA/SMS once), opens DevTools → Network, right-clicks any authenticated request → "Copy as cURL", and passes the command here. Returns a live session_id already carrying that site's cookies and sitting on the copied request's URL — the agent continues from where the human left off, no password or second login needed. Works with bash, PowerShell and cmd copy flavors.
Inputs: curl (required), account, use_proxy.
render_markdownAInspect
Render a markdown document into a deterministic, self-contained HTML artifact - the document layer, so the agent never writes HTML by hand. Prose rides a plain offline shell (no fonts, no scripts); archify fenced code blocks carry typed zero-coordinate diagram JSON (sequence, workflow, architecture, dataflow, lifecycle families) and render to inline SVG via the layout engine. Same input, same bytes: the receipt carries the sha256 so determinism is verifiable. theme picks light (default) or dark; preset picks the palette family — classic (default), signal-flow, blueprint, editorial — orthogonal to theme; colors bake at generation time (presentation attributes, not CSS variables), and the receipt records both preset and theme. quality picks the composition audit profile — standard (default) or showcase, the delivery gate: the receipt's diagrams[].composition grades route crossings, ambiguous corridors, label clearance (2px standard / 4px showcase), route rhythm, and node text projected to the 930px reader width; the audit never changes the artifact bytes. Mermaid sources are the agent's job to translate, not the engine's: flowchart/graph → workflow (lanes + columns), sequenceDiagram → sequence, stateDiagram-v2 → lifecycle (bands), erDiagram/class → architecture (grid + boundaries) — read the topology and emit the matching zero-coordinate archify JSON; the engine accepts only archify JSON. A broken diagram degrades to a visible code block and lands in receipt.diagnostics; an authored route preset that cannot be honored is self-repaired to a verified semantic substitute and disclosed in receipt diagrams[].repairs - the document still renders. A fence may also carry views: [{id,label,nodes,note?}] (node ids of the active family), emitted as guided-view tabs above the diagram plus an inlined viewer script - clicking a tab lights the member nodes and the routes between them (subgraph), clicking a node lights it with its direct neighbors (ego graph), everything else dims; a view's optional note shows as a caption while it is active (the story layer). window.agxViewer in a session drives and reads the same state programmatically: {focus,view,state} as before, plus route(i,from,to) which returns and lights the shortest authored directed path between two nodes (null when unreachable, state untouched), and reach(i,id,down|up) which returns and lights the authored downstream/upstream closure ({nodes,links}); both dim the rest of the diagram. diagrams[].views in the receipt lists the tabs. motion: true bakes an entrance choreography into the artifact: pure-declarative CSS animation with zero scripts - headings split into per-glyph (CJK) / per-word (latin) spans that rise in with expo easing, prose blocks stagger up an nth-child delay ladder, diagram figures grow in with a back ease (GSAP's easing math as public cubic-bezier equivalents, nothing embedded); the diagrams themselves play a flow story on the same clock - nodes land beat by beat, solid edges draw in (dash-offset), dashed returns fade, sequence messages arrive as sent - with a timed caption strip under each figure as the subtitles, which becomes a static transcript under prefers-reduced-motion; the file itself animates in any browser and the receipt records motion plus diagrams[].story (beat times and captions - the hook for muxing voice later). With session_id the artifact is also loaded into that session (local, free) and the reply carries viewport acceptance: scroll extents measured in the live session and graded fits/tall/wide/oversized, telling the agent how to read the page back. Diagram vocabulary adapted from archify (MIT).
Inputs: markdown (required), theme, motion, preset, +2 more.
render_pdfAInspect
Cut a rendered page into pages and package as PDF, PNGs, PPTX or DOCX. Print mode (no selector) paginates the document into fixed-height pages (default 794x1123, A4 @96dpi), breaking at top-level block boundaries — no half-cut text where a break can land on a block edge. Slides mode (selector set) makes one page per match, sized to that element — generate an HTML deck with one .slide per page and each becomes a deck page. format "pdf" (default) returns base64 image-based PDF; "png" returns one base64 PNG per page in pages_base64; "pptx" returns a base64 PPTX (one slide per page, deck-sized to the largest page); "docx" returns a base64 DOCX (one page-sized section per page, each section keeps its own height). Returns page count and packaging.
Inputs: url (required), width, format, height, +5 more.
render_videoAInspect
Render a page's animation timelines to an MP4 video. The page's scripts must expose window.__timelines — objects with duration() and pause(t) (a paused gsap.timeline registered there works as-is). Each frame seeks every timeline to t=i/fps and paints the viewport, so the output is deterministic — no wall clock in the pixel values. Audio: narration[] places TTS/voice clips at start times (mixed into one AAC track), audio adds looped background music, and subtitles_srt muxes an SRT as a soft mov_text track and (by default, burn_subtitles: false to opt out) burns the same cues into the frame pixels — QuickTime, WeChat and most social embeds ignore the soft track. Requires ffmpeg on the server. Returns base64 MP4 (H.264, yuv420p) plus frame count and durations.
Inputs: url (required), fps, audio, width, +10 more.
searchARead-onlyInspect
Search the web across Baidu/Bing/Sogou/WeChat/Google (aggregated + deduped) and optionally fetch the top results' full content. Use when the agent needs to FIND information online - replaces a search API. Supports image search returning direct image URLs. Optional engines: ["baidu"]-style filter by engine name (invalid names error with the valid list; /doctor lists them with live health). Optional time_range day/week/month/year for news freshness (engines without dated results ignore it). Response carries engine_errors explaining any engine that contributed nothing (CAPTCHA suspension, transient failure).
Inputs: q (required), engines, fetch_top, categories, +3 more.
session_challengesARead-onlyInspect
One-call risk-control report: did this session hit an anti-bot wall? Taobao/tmall's x5 risk control answers 200 like a normal response — either a redirect onto a punish page (tmd/punish, punish.taobao.com) or an MTop API body carrying FAIL_SYS_USER_VALIDATE / RGV587 / x5secdata. Returns {total, events:[{url,method,status,kind,via}]} where via says whether the wall was navigated into ("url") or swallowed by an API response ("body"). When there are hits, the response also carries the account name (which identity got walled) and a handoff instruction: the engine detects and surfaces but does not auto-bypass — a human opens the live view (/live?session= on the engine's HTTP port), solves the challenge in this session, and the retry rides the cookie that solving sets. Detection only; no automated solving or bypass.
Inputs: session_id (required).
session_clickAInspect
Click an interactive element by its index (from session_state output) inside a live browser session: scrolls it into view and fires a DOM click on the session's current page. Before clicking it re-verifies the element in the same frame — if the page changed since session_state (element detached, disabled, hidden, or covered by an overlay), it returns clicked:false with a reason ("detached"/"disabled"/"not_visible"/"covered_by") and, when covered, a covered_by description of the element that would eat the click — never a silent no-op. A submit click may navigate the session — the returned url/text_after reflect the page after the action, and session state (cookies, localStorage, globals) persists for follow-up calls. Indexes come from the most recent session_state; re-list after navigation.
Inputs: index (required), session_id (required).
session_click_xyAInspect
Click at viewport coordinates (CSS pixels) via real mouse events — pointerdown/mousedown, pointerup/mouseup, then click on whatever element is hit there. For canvas/map surfaces with no DOM element to index. click_count 2 adds dblclick.
Inputs: x (required), y (required), session_id (required), button, +1 more.
session_cloneAInspect
Derive a new browser session from a live one, carrying the full login state: cookies, localStorage/sessionStorage, viewport pin, dialog policy, proxy and keepalive flags. The source session stays untouched. Use to snapshot a logged-in state before risky actions, or to run the same login in parallel tabs. Returns {session_id (new), cloned_from, url, viewport}.
Inputs: session_id (required).
session_closeAInspect
Close a browser session and free its resources. For a persistent session this also drops the on-disk login snapshot - idle expiry keeps it, an explicit close does not.
Inputs: session_id (required).
session_consoleARead-onlyInspect
Read the session's recent page console output (log/info/warn/error) as {url, total, matched, messages:[{ts_ms, level, text, url}]}, newest last. Ring buffer of 500 entries; captures output from page scripts, clicks, evals and navigation alike. Optional filters: level (exact, e.g. "error"), since_ts (epoch ms), url_contains (page URL substring), limit (most recent N matches). The fastest way to see WHY a page misbehaves: click the button, call this, read the error.
Inputs: session_id (required), level, limit, since_ts, +1 more.
session_cookiesARead-onlyInspect
Export the session's current cookies as ["name=value", ...] for the page's URL. Use to persist a logged-in session and replay it later via session_create with cookies. Round-trips with session_create's cookies field.
Inputs: session_id (required).
session_createAInspect
Create a persistent interactive browser session for multi-step interaction - clicking, typing, scrolling, reading state across page transitions. Use when the agent must INTERACT with a page (login flows, forms, pagination, click-through) rather than read it once. Returns session_id; persists 8 min idle. With persistent:true the login state survives idle eviction and server restarts - the same session_id revives logged-in.
Inputs: url, width, height, mobile, +7 more.
session_dialogAInspect
Inspect or flip the session's dialog policy for window.alert/confirm/prompt. Dialogs never block the page: each is auto-answered (default dismiss) and logged into session_console at level "dialog". action "list" reports {policy, prompt_text, dialogs}; "accept" makes subsequent confirm() true and prompt() return prompt_text (or the call's default argument); "dismiss" restores the default.
Inputs: action (required), session_id (required), prompt_text.
session_dragAInspect
Drag the mouse from one viewport position to another: press at from, steps mousemove events, release at to. The trajectory is humanized by default (eased velocity, wobble, jittered timing, overshoot) — the shapes anti-bot checks score for; pass humanize:false for exact linear interpolation. Moves AMarker-style drag targets, canvas selections and captcha sliders that only track while the pointer travels.
Inputs: to (required), from (required), session_id (required), steps, +2 more.
session_evalAInspect
Execute arbitrary JavaScript in a live browser session and return the result. Runs in the session's current page, so DOM mutations, globals and storage persist across calls — unlike the stateless eval tool, which loads its own throwaway page each call. Script-driven navigation moves the session's URL. JS exceptions are reported with name, line/column and stack.
Inputs: script (required), session_id (required), timeout_ms.
session_exportARead-onlyInspect
Export a browser session's recorded action log. Format "bash" (default) returns a runnable curl script that replays every recorded action (navigate/click/input/scroll/eval) against a fresh session on this server — hand it to a shell or cron, zero model tokens. Format "jsonl" returns the raw action log, one JSON object per line. Format "json" returns a flow.json document — the same recording as editable ops ({op, args}) with cookies/storage stripped — that flow_run replays server-side.
Inputs: session_id (required), format.
session_inputAInspect
Type text into an input/textarea element by its index (from session_state output), focusing it and dispatching input/change events. events:"full" is the complete human typing gesture: per-character keydown/keypress/input/keyup cycles, trailing change, then blur — the tail blur commits on forms that save in onBlur (React capture listeners, #100). A disabled, readonly, or detached field answers filled:false with a reason instead of a silent write. Hidden inputs are legitimate targets and are filled normally.
Inputs: text (required), index (required), session_id (required), events.
session_listARead-onlyInspect
List live browser sessions with idle age and the time left before auto-eviction. Use to discover a session to reuse instead of creating a new one; sessions expire after 8 min idle.
Inputs: no named parameters.
session_networkARead-onlyInspect
Read the session's network request log. filter="media" extracts playback/stream URLs (m3u8/HLS, mp4, dash, flv...) actually requested by the page's player at runtime - the reliable way to get a real video link, since links embedded in page HTML are often decoys. Media elements and player iframes the engine never fetches (video/audio/source/iframe src) are merged in as candidates: via="network" entries are confirmed requests, via="dom" entries are candidates carrying their tag (iframes = kind "iframe", navigate into them to sniff). Default returns every request as compact rows (method/url/status/type/size). Navigate to the video page first, let it load, then call this.
Inputs: session_id (required), filter, url_contains, body_max_chars, +1 more.
session_preloadAInspect
Replace the session's document-start preload group (empty array clears). Sources run before each new document's own scripts — including inline tags — which is the only hook that beats pages whose signing layer captures window.fetch/XHR natives at parse time (xhs's inline jsvmp). Set before the first navigate; applies to every navigation from then on.
Inputs: session_id (required), scripts.
session_screenshotAInspect
Screenshot the session's CURRENT DOM state (mutations from clicks/evals included) as a base64 PNG via the built-in renderer. Width/height default to the session's viewport, so session_viewport + session_screenshot shows the responsive layout. Returns {url, width, height, image_base64, format}.
Inputs: session_id (required), width, height, selector, +2 more.
session_scrollAInspect
Scroll the page up or down by a number of viewport-heights.
Inputs: session_id (required), amount, direction.
session_set_filesAInspect
Select files on a file input programmatically (Playwright setInputFiles semantics): builds File objects from base64 content, assigns them to input.files, then dispatches input+change so framework onChange handlers fire. Selector-addressed because file inputs are often hidden and absent from the session_state index.
Inputs: files (required), selector (required), session_id (required).
session_stateARead-onlyInspect
Get the current page state as an indexed list of interactive elements. Returns compact text with [N] indexes for use with click/input tools.
Inputs: session_id (required).
session_storageARead-onlyInspect
Snapshot the session's localStorage/sessionStorage for the current origin: {url, local_storage, session_storage}. Feed it back via session_create's storage field to restore a logged-in state in a new session — the half of login state that cookies can't carry (many sites keep the session token in localStorage). Call before the session idles out.
Inputs: session_id (required).
session_verdictARead-onlyInspect
One call answers "where did this session land": verdict is one of challenge (risk control engaged — punish page or a 200-status API body that swallowed the wall; the response carries a handoff instruction for a human to solve it in the live view), captcha (explicit CAPTCHA interstitial), login (bounced to a login form — auth expired), empty, landed (normal 2xx content page), or unknown (couldn't classify — read facts; a non-2xx main document lands here with facts.doc_status carrying the number, so a zhihu-style burst 403 is branchable). The facts sheet also carries challenge_events, requests, console_errors and the fired signals. Pure code over signals the engine already holds (current URL, risk-control rows, main document status/size, console errors) — no screenshots, no page evals, single-digit milliseconds. Verdict observes, it never bypasses.
Inputs: session_id (required).
session_viewportAInspect
Set the session's viewport (device emulation): scripts see innerWidth/innerHeight move, media queries like (max-width: 600px) re-evaluate, element rects re-anchor, and mobile=true flips pointer/hover matchMedia answers to coarse/none. Omitted width/height keeps the current value.
Inputs: session_id (required), width, height, mobile.
session_waitARead-onlyInspect
Wait until a CSS selector matches or a JS predicate turns truthy, with a timeout. The page's event loop keeps running while waiting (fetches, timers, promise chains progress), so this replaces blind sleeps for async content: navigate, session_wait for '.price-card', then click/read. Returns {matched, elapsed_ms, detail:{tag,text} or the predicate value}; errors with timeout ... naming the selector/predicate on expiry. Exactly one of selector/predicate.
Inputs: session_id (required), selector, predicate, timeout_ms.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 40 tools
Tool families are cleanly separated: stateless fetch/eval/click are explicitly distinguished from their session_* counterparts, and account_*/render_*/session_* names signal intent. The main fuzzy boundaries are session_challenges vs session_verdict (both report anti-bot/risk status) and the three render_* tools, though descriptions usually resolve them.
Most tools follow predictable families—account_<verb>, session_<operation>, render_<format>—and bare stateless verbs (fetch, search, eval, click) are easy to read. Deviations such as flow_run instead of run_flow and noun-style read tools like session_state/session_network/session_cookies keep it from being perfectly uniform.
40 tools is a heavy surface, especially with 26 session_* variants; even though each is a real browser action, the set will tax an agent's selection and context budget. The server's broad scope explains the count, but it still exceeds what most MCP clients handle gracefully.
The surface covers the full browser workflow: read/search/download, session lifecycle and input, login-state persistence, risk-control verdicts, flow replay, and render/export targets. Minor gaps such as no explicit multi-tab/window management or standalone proxy/user-agent configuration are workaroundable.
Maintenance
Related MCP Connectors
Control a real browser from AI agents: navigate, click, type, screenshot, read the page.
Cloud browser, web extraction, screenshots, and remote CDP tools for AI agents.
Official Browserless MCP server: browser automation, scraping, screenshots and PDFs for AI agents.
Web infrastructure for AI agents: scrape, crawl and search the web, run browsers and agents.
Related MCP Servers
- AlicenseNot gradedqualityNot gradedmaintenanceA Playwright-based MCP server that exposes a live browser as a traceable, inspectable, debuggable and controllable execution environment for AI agents.5,213 npm57-
- AlicenseAqualityCmaintenanceBrowser automation MCP server that uses a real browser to give agents eyes and hands—open pages, click, fill, screenshot, and run scripts via accessibility-tree snapshots.226 npmMIT
- AlicenseNot gradedqualityAmaintenanceEnables an AI agent to observe, plan, and act inside a real browser session by exposing navigation, snapshot, click, type, and screenshot MCP tools over local JSON-RPC and SSE. It runs a Playwright-backed browser in-process and provides stable element references for reliable actions.MIT
- AlicenseNot gradedqualityAmaintenanceEnables AI agents to run Playwright-compatible browser automation with persistent profiles, proxies, Chrome extensions, and humanized mouse, keyboard, and scroll interactions over stdio or Streamable HTTP. Supports web research, daily automation, testing workflows, and optional session-scoped CDP access.280 npmMIT