接入你的智能体
從發布者安裝。不是 PlugMCP 端點。
遠端 URL 還是本機指令? 兩種安裝方式的差異
來自發布者
⚡ Temporal Browser Intelligence & Agent-Owned Workflow Runtime with 350 Certified MCP Tools & Python SDK · Sub-Millisecond DOM Time-Travel, Target Memory & Deterministic Automation.
TeleDOM is not the decision-maker; TeleDOM is the enabler.
The Agent is the brain. TeleDOM is the hands, eyes, memory and browser toolbox.
Teach the browser task once. Turn it into a reusable program. Run it anywhere the browser can go.
AGENT — discover · reason · write workflows · version · debug · repair · generate bots
│ MCP (Level 1: 350 tools) ──────────────┐
│ Python SDK (Level 2: semantic browser) ─┐ │
▼ ▼
TELEDOM — browser observation · primitives · target memory ·
workflow persistence · DUMB execution · evidence · replay · proof
RUN #1 (explore) 5 tool calls · 2 DOM scans · 5 MCP round trips
→ agent learns: saves learned targets + writes the workflow
RUN #2 (reuse) 1 td_workflow_run call · 1 DOM scan
✔ verify_cta ✔ click_cta ✔ extract ✔ assert → SUCCESS
deterministic execution record + verbatim replay
KPIs: 80% fewer MCP round trips · 50% fewer DOM scans · replay PASS
0 unsafe-action bypass · 0 workflow corruption · 350/350 certified
Browser-first, API-optional: web applications and complex SPAs without public APIs are the primary target, not the exception. The AI agent chooses its own discovery strategy (DOM → semantics → accessibility → script → coordinates) whenever abstractions fail — TeleDOM provides rock-solid, deterministic execution primitives.
TeleDOM in one line: See what happened. Understand why. Simulate what-if. Fix safely. Prove the result.
TeleDOM is a production-grade Temporal Browser Intelligence Engine and universal Model Context Protocol (MCP) platform providing 350 certified tools for autonomous AI coding agents (Claude, Cursor, Antigravity, Cline, OpenAI Swarm) and frontend engineering teams.
Every important claim is backed by mathematical evidence, confidence scoring, provenance tracking, and machine-verifiable proof.
npm run bench — full matrix)
🎬 Live Automation Demo: Autonomous AI Coding Agent driving real-time browser forensic recording, live DOM inspection, synthetic actions, and multi-turn autonomous web flows. (Watch Full 1080p Video)
dom-antigravity)TeleDOM provides an official, zero-dependency Python SDK (sdk/python/teledom):
from teledom import Browser, Workflow
with Browser() as browser:
# 1. High-level browser control
page = browser.inspect()
browser.type("#search-box", "Autonomous Agents")
browser.click("#submit-btn")
# 2. Workflow creation & execution
wf = Workflow("triage_issues", client=browser.client)
wf.input("filter", "Issue label filter", default="bug")
wf.step("filter_step", "td_target_check", args={"selector": ".issue-row"})
wf.save(version="1.0.0")
# 3. Execute with deterministic audit record
run = wf.run({"filter": "security"})
print("Workflow run status:", run["run"]["status"])
dom-antigravity)The package exposes a global CLI binary registered as dom-antigravity (with aliases mcp-dom and browser-antigravity):
# 1. Install MCP configuration & skills into current workspace (.agents)
dom-antigravity install --workspace
# Short alias:
dom-antigravity install -w
# 2. Install globally for ALL projects in Antigravity IDE
dom-antigravity install --global
# Short alias:
dom-antigravity install -g
# 3. Install into a specific target directory
dom-antigravity install --target "C:/path/to/project"
# 4. Check bridge server health & connected Chrome tabs
dom-antigravity status
# 5. Capture clean live Chrome screenshot (Full Page + Cropped Element)
dom-antigravity screenshot
# 6. Start WebSocket Bridge manually (Optional — Auto-Bridge handles this on demand)
dom-antigravity bridge
# 7. Print ready-to-use JSON configuration for Claude Desktop or Cursor
dom-antigravity config cursor
dom-antigravity config claude
TeleDOM operates cleanly across two distinct runtime environments:
Browser Runtime (Chrome Extension Manifest V3 / Content Script / Injected Page Context):
MutationObserver.LogicalNodeId identifiers to live nodes via WeakMap.Ctrl + Shift + Mouse Click for visual element picking.click, type, hover, focus, scroll, drag).Node.js / MCP Server Runtime (ForensicMCPServer):
stdio and HTTP.3847.┌────────────────────────────────────────────────────────────────────────┐
│ AI AGENT / MCP CLIENT │
│ (Antigravity IDE / Cursor / Claude Desktop / CLI) │
└───────────────────────────────────┬────────────────────────────────────┘
│ (JSON-RPC 2.0 over stdio)
▼
┌────────────────────────────────────────────────────────────────────────┐
│ TELEDOM V4.1 MCP SERVER (350 TOOLS) │
│ ├── 144 td_* Temporal Intelligence & Workflow Runtime Tools │
│ ├── 54 dt_* Chrome DevTools Fusion Tools │
│ ├── 31 fx_* Advanced Forensic Capabilities │
│ ├── 121 Core & Live Interaction Tools │
│ └── Zero-Config On-Demand Auto-Bridge Dispatcher │
└───────────────────┬────────────────────────────────┬───────────────────┘
│ │ (WebSocket / HTTP :3847)
▼ ▼
┌───────────────────────────────┐ ┌────────────────────────────────────┐
│ WORKFLOW RUNTIME │ │ LIVE WEBSOCKET BRIDGE SERVER │
│ ├── Workflow Persistence │ │ (Port 3847 - Auto-Managed) │
│ ├── Target Memory Store │ └─────────────────┬──────────────────┘
│ ├── EventMesh & Causal Graph │ │
│ ├── Counterfactual Engine │ │ (Bi-directional JSON frames)
│ └── Verification & Proofs │ ▼
└───────────────────────────────┘ ┌────────────────────────────────────┐
│ CHROME EXTENSION (MV3) │
│ ├── Mutation Observer Engine │
│ ├── Ctrl+Shift+Click Live Picker │
│ ├── Canvas Screenshot Pipeline │
│ └── Synthetic Interaction Driver │
└────────────────────────────────────┘
teledom/
├── src/
│ ├── intelligence/ # Temporal Intelligence Engine (EventMesh, Causal, Proof, Incident)
│ │ └── workflow/ # v4.1 Agent-Owned Workflow Runtime (domain, executor, store, facade)
│ ├── core/ # Core recorders, sequence counters, privacy, PNG builder
│ ├── diff/ # Structural DOM diff engine (attributes, classes, styles, subtrees)
│ ├── extension/ # Chrome Extension Manifest V3 (content scripts, service worker)
│ ├── lifecycle/ # Lifecycle tracer, disappearing UI analyzer
│ ├── mcp/ # Universal MCP Server, 350 tools definition, dispatchers
│ ├── reconstruction/ # Sub-millisecond snapshot interpolation & time-travel
│ ├── storage/ # Local disk storage & indexing engine
│ └── ui/ # Observatory UI, visual state viewers
├── sdk/
│ └── python/ # Official Python SDK (BrowserController, WorkflowEngine, TargetMemory)
├── docs/
│ ├── TOOLS_CATALOG_350_EN.md # Complete English reference for all 350 tools
│ ├── TOOLS_CATALOG_350_FA.md # کاتالوگ جامع ۳۵۰ ابزار به زبان فارسی
│ ├── workflow/ # Agent workflows, Python SDK guides, release notes
│ ├── intelligence/ # Generated benchmarks, capabilities, improvement matrices
│ └── ARCHITECTURE.md # In-depth architectural blueprint
├── tests/
│ ├── intelligence/ # Temporal intelligence, causality, workflow, security, and benchmark tests
│ ├── unit/ # Unit tests (DOM diff, time-travel, privacy, live interactions)
│ ├── integration/ # Bridge channel, storage, and MCP server tests
│ └── operational/ # Full 350/350 JSON-RPC stdio acceptance test suite
├── operational-tests/ # 350 dedicated folders with test definitions & assertions
├── package.json # Version 4.1.0, scripts, dependencies
└── README.md # English documentation (this file)
TeleDOM v4.1 exposes 350 production-ready JSON-RPC 2.0 tools grouped into four distinct families.
For detailed schemas, parameter documentation, and operational examples for all 350 tools:
Install the global CLI and register TeleDOM into your user configuration:
npm install -g teledom
dom-antigravity install --global
To enable TeleDOM for an individual project repository:
cd /path/to/your-project
npx teledom install --workspace
This creates .agents/mcp_config.json and registers the full 350-tool skill definitions into .agents/skills/.
chrome://extensions/.teledom/dist/extension folder.claude_desktop_config.json){
"mcpServers": {
"teledom": {
"command": "node",
"args": ["C:/path/to/teledom/dist/server/mcp-server.js"]
}
}
}
.cursor/mcp.json){
"mcpServers": {
"teledom": {
"command": "node",
"args": ["C:/path/to/teledom/dist/server/mcp-server.js"]
}
}
}
TeleDOM is backed by a rigorous multi-tier testing pipeline:
# 1. Run full unit and intelligence test suite (319 tests)
npm run test:unit
# 2. Run Python SDK self-test (17 tests)
npm run test:sdk
# 3. Run full operational stdio JSON-RPC certification (350 tools)
npm run test:operational
# 4. Run deterministic benchmark matrix (10K / 100K / 1M events)
npm run bench
# 5. Run golden incident suite (1,032 scenarios)
npm run golden
# 6. Run chaos engineering injection suite
npm run chaos
# 7. Build production bundle (client, server, extension)
npm run build
PrivacyEngine automatically sanitizes password fields, credit card numbers (Luhn-compliant), Social Security numbers, bearer tokens, API keys, and custom CSS selectors in both recording and live inspection.on* event handlers and prevent script execution, eliminating arbitrary code execution risks.requestAnimationFrame and requestIdleCallback throttling to ensure zero dropped frames on 120 Hz displays.npm run bridge in a separate terminal?A: No!
ForensicMCPServerincludes a zero-config auto-bridge. Whenever an AI agent connects via MCP, the server automatically initializes the WebSocket bridge on port3847in the background.
A: Run
dom-antigravity screenshot. It connects to your active Chrome tab, captures a pristine full-page screenshot and cropped element screenshot, and saves them to disk.
A: Every single tool is documented with full parameter schemas, TypeScript interfaces, and concrete operational JSON-RPC examples in TOOLS_CATALOG_350_EN.md and TOOLS_CATALOG_350_FA.md.
Licensed under the Apache License, Version 2.0. See the LICENSE file for complete terms:
http://www.apache.org/licenses/LICENSE-2.0
⚡ Temporal Browser Intelligence & Agent-Owned Workflow Runtime with 350 Certified MCP Tools & Python SDK · Sub-Millisecond DOM Time-Travel, Target Memory & Deterministic Automation.
工具
尚無工具清單。遠端伺服器會在首次健康檢查後取得工具清單。
常見問題
在本頁複製適用於 Cursor、Claude Desktop、Claude Code 或 VS Code 的 stdio 設定。它使用發布者的安裝方式,而不是 PlugMCP 代理伺服器。
這筆記錄未列出必要的環境變數。
TeleDOM 透過 stdio 在本機執行。
上次檢查時間:2026-10-28。 ⚡ Temporal Browser Intelligence & Agent-Owned Workflow Runtime with 350 Certified MCP Tools & Python SDK · Sub-Millisecond DOM Time-Travel, Target Memory & Deterministic Automation.