Skip to main content
Glama

⚡ Thunder

The human engine for AI agents.

Give any software a human body: a real browser with a real fingerprint, a mouse that trembles like a hand, keystrokes at human cadence, and a TLS handshake indistinguishable from Chrome's.

A bot that walks like a person.

PyPI Python License: MIT Tests Fingerprint

Install · Watch the proof · MCP for any software


Why Thunder exists

Every anti-bot system on Earth looks for the same three tells:

  1. Protocol tells — a Python scraper's TLS/HTTP-2 handshake screams "bot" before a byte of JS runs.

  2. Fingerprint tells — navigator.webdriver, headless quirks, software-rendered WebGL, empty plugin arrays.

  3. Behavioral tells — perfectly straight mouse paths, instant typing, zero reading pauses.

Thunder is a body for your AI that has none of them. Not patched over — absent, the way a real human's browser is.

pip install thunder-agent
import thunder

body = thunder.Body(profile="ada")          # a persistent human named Ada
await body.warmup()                         # age the session like a returning user
await body.go("https://protected-site.com")
await body.type("#email", "ada@example.com")   # typed like a person. Typos included.
await body.look("proof.png")                    # screenshot evidence
await body.rest()                              # close, keep the identity

Ada comes back tomorrow with the same cookies, same screen, same GPU string, same timezone. The site sees a returning regular.


Related MCP server: Real Browser MCP

📦 Install in 30 seconds

pip install thunder-agent

Works on Linux, macOS, Windows. Real-Chrome engine auto-detected; falls back to bundled Chromium.

thunder demo        # watch it browse like a human, right now
thunder ghost https://nowsecure.nl    # TLS-level fetch that passes the gate

⚡ The five layers

Layer

What it does

The tell it kills

Body

Stealth browser: real Google Chrome, persistent profiles

headless shells, disposable identities

Skin

One consistent identity: UA, platform, GPU, screen, timezone, locale

fingerprint mismatches, SwiftShader WebGL

Hand

Bézier mouse with hand tremor, human keystroke cadence, typos it fixes

straight-line paths, machine-perfect timing

Ghost

Pure HTTP with Chrome's exact TLS/JA3 fingerprint + full header orchestration + human-style retries

Python-scrape handshake = instant 403

Pool

Thunder's own proxy pool: auto-fetched, Ghost-validated egress, sticky per profile

shared/flagged datacenter IPs

Warmup

Organic light browsing before your target

cold first-hit sessions

Verified live: bot.sannysoft.com under Thunder — WebDriver missing (passed) · WebDriver Advanced passed · Chrome present (passed) · PluginArray passed — 4 passed, 0 failed with headed real Chrome. Ghost pulls a 200 from a Cloudflare gate that 403s plain Python, same minute.


🔌 MCP: give ANY software a human body

Run the full Thunder app anywhere — one command, no account needed:

pip install thunder-agent[app]
thunder app

Thunder speaks: the Mouth gives it a real voice — Piper TTS as the primary offline engine (matured male voice) and Kokoro-82M for premium-quality narration, with edge-tts (natural male Christopher voice) as the online fallback and the default on small hosted instances where local engines would exhaust memory. During any mission you can hear Thunder narrate each step as it works, and the 🎙️ button lets you speak your mission instead of typing it.

The Thunder app is a desktop-grade web control room: mission control, a live monitor streaming real Chrome as it works, Autopilot (the Brain — StewMind 1.7b, falling back to Groq and Pollinations — plans each step and Thunder executes it on camera), and Recordings: every session's frames and step log are saved automatically, so you can replay exactly how Thunder navigated and did the task. With FIREBASE_API_KEY set it gains full accounts + per-user MCP tokens + cloud history; without it, it runs in local zero-setup mode.

Hosted MCP endpoint (connector URL for Claude): https://thunder-app-s94i.onrender.com/mcp — hosted on the live Thunder App (https://thunder-app-s94i.onrender.com). The same server serves the desktop-style web app, accounts, live video and recordings. You get thunder_search, thunder_read, thunder_ghost, thunder_feel, thunder_solve, thunder_open — search, clean extraction and bot-gate-safe fetches through Chrome's exact TLS handshake, from inside Claude.

Want the full human body (stealth Chrome, mouse tremor, human typing) under MCP? Run it locally — 15 tools, full parity with the hosted endpoint plus the browser-only ones:

Thunder ships an MCP server (15 tools). Paste one config and Claude, Cursor, Cline, Windsurf — anything that speaks MCP — browses as a human:

{ "mcpServers": { "thunder": {
    "command": "pipx", "args": ["run", "thunder-agent", "thunder-mcp"] } } }
{ "mcpServers": { "thunder": {
    "command": "uvx", "args": ["--from", "thunder-agent", "thunder-mcp"] } } }

Local tools (full set): `thunder_open`, `thunder_read`, `thunder_click`, `thunder_type`, `thunder_scroll`, `thunder_look`, `thunder_ghost`, `thunder_search`, `thunder_feel`, `thunder_captcha`, `thunder_challenge`, `thunder_human_read`, `thunder_identity`, `thunder_think`, `thunder_rest`. Hosted endpoint tools: `thunder_search`, `thunder_read`, `thunder_ghost`, `thunder_feel`, `thunder_solve`, `thunder_open` (guide for running the full body locally). Same core, same code — online and offline behave identically.

Tools: thunder_open · thunder_read · thunder_search · thunder_click · thunder_type · thunder_scroll · thunder_ghost · thunder_solve · thunder_feel · thunder_close — each one moves like a person, never like a script.


🚫 The blocker table — what used to stop bots, and Thunder's answer

Blocker

Thunder's counter

Datacenter IP suspicion

ProxyManager: sticky per-profile residential proxies, health-checked, auto-rotation — THUNDER_PROXY env or ProxyManager.from_env()

No GPU / software WebGL

Full WebGL1 and WebGL2 spoof: vendor + renderer read as a real consumer GPU, on every API path detectors use

Headless leaks

Headless-evasion pack: outerWidth/Height, colorDepth, broken-image dimensions, full window.chrome shape, battery API

Advanced CAPTCHAs

Keyless soft_solve (human-hand click inside challenges) + ocr_solve (local OCR for image/math) + CapSolver/2Captcha bridge for the hardest

Behavioral ML

Hand tremor + micro-corrections + warmup sessions + thinking pauses — entropy that reads as a person, because it moves like one

🔒 CAPTCHAs without a subscription

from thunder.captcha import soft_solve, ocr_solve
result = await soft_solve(body.page)     # human-hand click, no keys needed

Non-interactive challenges often pass on a warmed-up session. For Turnstile/reCAPTCHA Enterprise, wire your own CapSolver or 2Captcha key — Thunder senses the challenge type and routes it.

v0.5 — ChallengeResolver, Thunder's own bypass system (no external deps):

from thunder.captcha import ChallengeResolver
verdict = await ChallengeResolver().resolve(body.page, body=body)
# {'solved': True, 'kind': 'turnstile', 'how': 'human-timed checkbox click'}

Detects the challenge type — Turnstile, reCAPTCHA v2, hCaptcha, Amazon's text captcha, Cloudflare interstitials — and runs the built-in counter: hand-timed checkbox clicks, patience for interstitials, local OCR for Amazon (retrying like a person squinting). No paid keys, no subscriptions.


🫆 Wear YOUR OWN fingerprint

The headline feature of v0.5. Open https://browserleaks.com in your normal browser, copy your values, and Thunder wears your machine's ghost — same UA, GPU strings, screen, cores, RAM, timezone, languages:

from thunder.fingerprint import Skin
from thunder.body import Body

me = Skin.imitate({
    "user agent": "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) ...",
    "gpu renderer": "ANGLE (Apple, ANGLE Metal Renderer: Apple M2, ...)",
    "timezone": "Europe/London", "screen": "2560x1440",
})
body = Body(skin=me)            # your machine's ghost

Or from the terminal:

thunder identity --out my.json        # export the live fingerprint
thunder identity --imitate my.json    # wear it and verify live

Partial data is fine — anything missing stays Thunder-consistent. An imitated skin is mutually consistent: your Mac UA will never claim a Windows GPU.


📖 Reads like a human too

await body.read_like_human(seconds=15)   # burst-scroll, reading pauses,
                                         # idle drift, re-reads

Behavioral ML scores dwell patterns. Thunder gives pages a genuine reading signature — pauses at paragraph speed, hands that drift while "reading", occasional scroll-backs.


🧠 Pair with any mind

Thunder is the body; bring any brain. Works with stewmind, Ollama, LM Studio, vLLM, OpenAI-compatible APIs, or fully offline transformers.

from thunder.mind import Mind
mind = Mind(provider="ollama", model="llama3")
plan = await mind.think("find the pricing page and extract the tiers", body)

🧪 Tested, not promised

git clone https://github.com/emmanuelerogian723-alt/thunder
cd thunder && pip install -e ".[dev]" && pytest

Stealth regression suite: webdriver erased + scan-invisible · genuine PluginArray · headless pack · WebGL skin-consistency · proxy rotation. Each one fails loudly if a future commit breaks the human illusion.


⚖️ The honest limits (why people trust Thunder)

  • No tool on Earth is "undetectable." Anyone selling that is lying to you.

  • Cloud IPs still look like clouds — pair Thunder with a residential proxy for the hardest targets.

  • Google-scale behavioral ML scores the whole session. Thunder's warmup + hand give you a real chance; it's not a magic key.

  • Use Thunder only where you're authorized. Your accounts, your tests, your sites. That's the license, and it's the line.

MIT licensed. Built in the open — every claim on this page has a script in tests/ or a screenshot in proofs/.


Watch Thunder beat the bot-detection gauntlet, live:

https://emmanuelerogian723-alt.github.io/thunder/

⭐ If this saved you a month of anti-detect work, star the repo — it helps others find it.

Thunder — built so software can finally act human. 🌩️

Available Tools

15 tools
thunder_captchaB

Detect CAPTCHA challenges on the current page.

Inputs: no named parameters.

thunder_challengeB

Detect and counter any CAPTCHA/challenge on the current page (Turnstile/reCAPTCHA/hCaptcha/Amazon/CF interstitial) — keyless.

Inputs: no named parameters.

thunder_clickC

Click like a human (bezier mouse path + human timing).

Inputs: selector (required).

thunder_feelC

Probe a URL's bot posture: plain request vs thunder ghost (TLS-true) request.

Inputs: url (required).

thunder_ghostB

Fetch a URL with a real browser TLS fingerprint (curl_cffi).

Inputs: url (required).

thunder_human_readC

Read the current page like a human: burst scroll, reading pauses, idle hand drift.

Inputs: seconds.

thunder_identityB

Export the live fingerprint, or wear one from a fingerprint JSON file (Skin.imitate).

Inputs: imitate_file.

thunder_lookA

Take a screenshot of the current page; returns the file path.

Inputs: no named parameters.

thunder_openC

Open a URL in Thunder's human-shaped browser body.

Inputs: url (required).

thunder_readB

Read the current page: title, url, text, clickable elements.

Inputs: no named parameters.

thunder_restB

Close the body. Cookies/fingerprint persist for next time.

Inputs: no named parameters.

thunder_scrollC

Scroll with human easing and reading pauses.

Inputs: amount.

thunder_thinkC

Ask stewmind-1.7b for the next action toward a goal.

Inputs: goal (required).

thunder_typeB

Type like a human (inter-key delays, rare typos, pauses).

Inputs: text (required), selector (required).

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    B3/5.0

    Scored across 15 tools

    Disambiguation3/5

    Most tools are distinct, but several pairs overlap: thunder_read vs thunder_human_read, thunder_captcha vs thunder_challenge, and thunder_ghost/thunder_feel/thunder_search all involve browser-TLS fetching. thunder_rest is especially unclear as a name for closing the browser body.

    Naming Consistency3/5

    The thunder_ prefix is consistent, but suffixes mix verbs (open, read, click) with nouns (captcha, identity, challenge, ghost) and compounds (human_read). Some names like thunder_rest and thunder_feel are cryptic about what they actually do.

    Tool Count4/5

    15 tools is reasonable for a stealth browser automation server covering navigation, interaction, fingerprinting, CAPTCHA handling, search, and planning. There is slight redundancy in the CAPTCHA and page-reading tools, but the count is not excessive.

    Completeness4/5

    Core browser lifecycle is covered: open, read, click, type, scroll, screenshot, and close. It also adds fingerprinting, search, CAPTCHA handling, and a reasoning helper; minor gaps like tab management, navigation history, or explicit waits are workaroundable.

    Maintenance

    ActivityMaintained
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      D
      maintenance
      Provides a real browser that bypasses bot detection (Cloudflare, Turnstile) for AI agents, enabling navigation, clicking, typing, screenshots, and data collection through MCP tools.
      166 npm
      MIT
    • A
      license
      Not graded
      quality
      A
      maintenance
      Enables AI agents to automate a real Chromium browser through the unchanged upstream Playwright MCP tool surface, covering navigation, research, testing, and daily web workflows in plain language. It ships via npm or Docker over stdio or Streamable HTTP, and supports persistent login profiles, Chrome extension loading, proxy/GeoIP matching, and humanized mouse, keyboard, and scroll behavior.
      431 npm
      MIT
    • A
      license
      Not graded
      quality
      A
      maintenance
      Enables MCP-compatible AI agents to drive an anti-detection browser for navigation, interaction, search, extraction, downloads, and persistent multi-tab sessions across 47 tools. Connects over stdio or HTTP to a separately running browser server, letting agents automate the web without getting blocked.
      899 npm
      MIT