Hermes Browser MCP Server
Allows AI agents to log into Gmail, retrieve unread emails, and perform other email-related actions through browser automation.
Supports browsing and interacting with Google services, including login and credential management for Google sites.
Enables AI agents to navigate, search, and interact with Naver services, including performing searches and managing credentials via browser automation.
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Hermes Browser MCP ServerGo to google.com and search for the weather today"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Hermes Browser V12
AI-powered Electron browser with MCP-First architecture for engineers.
Status (2026-07-11): V12 complete. 157 unit PASS + 46/46 eval + 49 MCP tools + 10 LLM providers + Cowork for engineers. Real production-ready.
π― V12 Three Differentiators
1. MCP-First Architecture
Everything exposed as MCP tools (49 total). External AI agents (Claude, GPT, Hermes AI, etc.) can fully control the browser via HTTP bridge or stdio server.
Browser navigation, click, fill, autofill, search
Content extraction: page text, tables, forms, links, search results
File management: list, read, search, grep, stat (Cowork)
Network: search Naver, search DDG, extract via URL
Vision: prints PDF, downloads, find text in page
Multimedia: browse screenshots, capture form fields2. Local-First + BYOK
Local LLM support + 10 cloud providers. No provider lock-in. BYOK (Bring Your Own Key) for all cloud providers.
Provider | Type | Use case |
mock | OpenAI-compat | Local mock (opencode-go proxy) |
LM Studio | OpenAI-compat | Local inference (:1234) |
Ollama | OpenAI-compat | Local inference (:11434) |
OpenAI | OpenAI-compat | gpt-4o-mini default |
Anthropic | Native | Claude 3.5 Haiku |
Native REST | Gemini 2.5 Flash | |
OpenRouter | OpenAI-compat | Aggregator |
MiniMax | Anthropic-compat | M3 |
BrowserOS | OpenAI-compat | Local fork (:8765) |
OpenAI-compatible | OpenAI-compat | Custom URL |
3. Cowork for Engineers
Files + browser + AI ν΅ν©. BrowserOS Coworkμ κ°ν λ²μ β νλ‘ λλ©μΈ νΉν.
cowork_list β list files by pattern
cowork_read β read text (max 5MB, binary β metadata)
cowork_grep β regex search across files (cross-platform)
cowork_search β by name OR content pattern
cowork_stat β metadata (size, mtime, mime)Default workspace: C:\Users\qqwer\Hermes-Workspace\
Related MCP server: BrowserMCP Secure
π¦ 30-Second Quick Start
Test & Run
cd /mnt/c/Users/qqwer/Desktop/Hermes/hermes-browser
npm test # 157 unit + 46 eval PASS (~3 sec)
npm start # Launch Electron (Miraecle window)Live Verification
Hermes Browser auto-starts HTTP bridge on port 8780. Connect from external AI agents:
TOKEN=$(curl -s http://127.0.0.1:8780/auth/token | jq -r .token)
# Call any of 49 tools
curl -X POST http://127.0.0.1:8780/mcp/tool \
-H "Authorization: Bearer $TOKEN" \
-d '{"name":"cowork_list","args":{"dir":"demo-circuits"}}'Use with Claude Code / Cursor / Cline
Add to ~/.claude.json (stdλ‘ MCP μ¬μ©):
{
"mcpServers": {
"hermes-browser": {
"command": "node",
"args": ["C:/Users/qqwer/Desktop/Hermes/hermes-browser/mcp-server/server.js"]
}
}
}ποΈ Architecture (V12)
ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β Renderer (src/chrome.html) β
β - Tabs, sidebar, chat UI, settings popover β
β - Light/dark theme (Space Grotesk + DM Sans) β
β - Spring motion (--ease-spring) β
ββββββββ¬ββββββββββββββββββββββββββββββββββββββ¬ββββββββββββ
β IPC β
β β
ββββββββΌββββββββββββββ βββββββββββββΌβββββββββββ
β Main Process β β MCP Bridge β
β (main.js) βββββββββββββΊβ (HTTP :8780) β
β - nativeTheme β β 49 tools exposed β
β - syncNativeTh.. β β localhost only β
β - Cowork attach β β Token auth β
ββββββββ¬ββββββββββββββ ββββββββββββββββββββββββ
β
ββββββββΌββββββββββββββββββββββ
β Agent Service β
β (src/agent/index.js) β
β - Plan/Mode/Approval β
β - ReadingList/Workspace β
β - Session Recorder β
β - CredentialVault β
β - CoworkService (V12) β
β - Provider abstraction β
β - Scheduler β
ββββββββββββββββββββββββββββββKey Files
File | Purpose |
| Electron main, theme sync, IPC, MCP bridge attach |
| UI shell with glass + light/dark + spring motion |
| AgentService (single API entry point) |
| V12 Cowork: files + browser + AI integration |
| HTTP server exposing 49 tools on port 8780 |
| 6 unit tests for cross-platform path handling |
| 157 unit tests total |
| 46 integration scenarios |
π V12 Features (Day 1-5)
Day 1: OS Theme Sync
nativeTheme.themeSource = 'system'OS λ³κ²½ μ chrome.html
data-theme="dark"μλ μ νκ²μ¦: light 100%
#f5f5f7, dark 100%#121212
Day 2: BYOK + 10 LLM Providers
Provider presets μλ μ±μ°κΈ° (URL/model)
3 endpoint νμ dispatch (OpenAI-compat / Anthropic native / Google native)
λΌμ΄λΈ κ²μ¦: λͺ¨λ 10κ° provider endpoint λλ¬
Day 3: MiniMax Endpoint Fix
base_url =
https://api.minimax.io/anthropicendpoint =
{base}/v1/messagesHeader =
X-Api-Key(capitalized)κ²μ¦: 401 "login fail: Please carry the API secret key" (endpoint OK)
Day 4: Cowork + 49 Tools + UI Spring Motion
src/agent/cowork.js240+ lines (V12 day 1)5 public Cowork methods + 13 V12 tools (browser/web extensions)
--ease-spring: cubic-bezier(0.34, 1.56, 0.64, 1)+ 7 keyframesSpring-in/slide/sidebar/chat/pulse/focus animations
Day 5: Cross-Platform Path Fix + νλ‘ Demo
_safePathcross-platform (Windows path β WSL β/mnt/...)Electron path λͺ¨λ (Windows mode) μλ
BD69730FV νλ‘ demo (BOM 18 parts, Gerber, datasheet)
5/5 cowork e2e ν΅κ³Ό
π οΈ Build & Verification
# Run all tests
npm test
# Quick unit-only run
node tests/agent.test.js
node tests/cowork.test.js
# Eval scenarios (46 total)
node eval/runner.js
# Live MCP call examples
curl -s http://127.0.0.1:8780/mcp/tools | jq '.tools | length'
# Open live window
npm startπ License
AGPL-3.0 (no BrowserOS code copy, safeStorage-based credential vault)
π¨ Design Choices
Light theme:
#f5f5f7background, white cards + box-shadow, gold#fbbf24accentDark theme:
#0f0f1ebackground, glass black 0.72 + blur 24pxFonts: Space Grotesk + DM Sans
Spring motion: tagu requested "μ«λν" easing β
cubic-bezier(0.34, 1.56, 0.64, 1)Glass:
backdrop-filter: blur(24px)for panels/popovers
π€ Acknowledgements
Inspired by browserOS-ai (AGPL-3) for the MCP-first + cowork architecture. Safe fork β no code copied, all implementation original.
This server cannot be installed
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Related MCP Servers
- Alicense-qualityDmaintenanceEnables real browser automation as tools in Cursor, Claude Desktop, Windsurf, and any MCP-compatible client, allowing AI agents to interact with web pages through natural language.Last updated36MIT
- Alicense-qualityCmaintenanceSecurity-hardened MCP server that gives AI assistants full control over your real browser session, supporting 36 tools for navigation, data extraction, monitoring, and more.Last updatedMIT
- Flicense-qualityDmaintenanceEnables AI to control browsers via natural language for web automation, testing, and data scraping. Supports Chrome-based browsers and integrates with any MCP-compatible AI tool.Last updated2
- Flicense-qualityDmaintenanceA browser automation MCP server providing 30 tools for navigation, interaction, page information, state checks, tab management, and more, enabling natural language control of browsers via MCP-compatible clients.Last updated
Related MCP Connectors
Browser MCP for logged-in tasks. Uses your Chrome β credentials stay local. Zero-token replay.
Live browser debugging for AI assistants β DOM, console, network via MCP.
A paid remote MCP for AI agent browser MCP session, built to return verdicts, receipts, usage logs,
Latest Blog Posts
- Who's Calling? MCP Hosts Are an Identity Blind Spot (And the Spec Knows It)By Om-Shree-0709 on .mcpAgent IdentityOAuth 2.1
- Your AI Chatbot Just Exposed Your CEO's Salary to an InternBy Om-Shree-0709 on .Agent IdentityMCP SecurityOAuth Delegation
- Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)By Om-Shree-0709 on .Agentic AiPrompt InjectionWebAssembly
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/leemind-q/hermes-browser'
If you have feedback or need assistance with the MCP directory API, please join our Discord server