Global WebMCP
Adds a versioned action for searching stays on Airbnb by destination, check-in/close, and guest counts; current execution requires a browser session.
Catalog includes a versioned Amazon UK product search action, currently blocked from execution.
Adds a versioned action to get a Facebook profile; current execution requires a browser session.
Provides executable GitHub search actions and includes a blocked GitHub issue action.
Adds a versioned action to search the web on Google; current execution requires a browser session.
Adds a versioned action to get an Instagram profile; current execution requires a browser session.
Adds a versioned action to search Reddit posts; current execution requires a browser session and CAPTCHA was observed.
Adds an executable versioned action to list 'for you' videos from TikTok via public GET.
Adds a versioned action to send a WhatsApp message, but write actions and browser sessions are blocked.
Provides executable versioned actions for searching Wikipedia articles.
Adds a versioned action to search YouTube videos; current execution requires a browser session.
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., "@Global WebMCPRun the Wikipedia search action for 'MCP'."
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.
Browser agents can discover how a website works. Making them rediscover the same workflow on every run is slow, expensive, and brittle.
Global WebMCP separates discovery from execution. Codex drives the website, measures the result, and writes a strict versioned action map. A local MCP server turns that map into a persistent function call.
A click trace is evidence, not an API. The map stores the narrowest stable interface the agent found.
See it work
Ask the catalog for a Wikipedia action:
{
"tool": "find_sites",
"arguments": {
"query": "search Wikipedia"
}
}The catalog returns a versioned reference:
en.wikipedia.org/search_articles@1.1.0Call it:
{
"tool": "execute_site_action",
"arguments": {
"actionRef": "en.wikipedia.org/search_articles@1.1.0",
"input": {
"query": "web model context protocol"
}
}
}The live verifier gets HTTP 200 and returns:
[
"AT Protocol",
"HTTP",
"Communication protocol",
"HTTPS"
]Run the same proof yourself:
npm run verify:wikipediaRelated MCP server: Sophia Stack MCP
Install
Global WebMCP is a local stdio server. It needs Node.js 20.19 or newer.
git clone https://github.com/benjang032/globalwebmcp.git
cd globalwebmcp
npm install
npm run buildAdd it to Codex from the repository root:
codex mcp add globalwebmcp -- node "$PWD/dist/src/index.js"For another MCP client, configure the same process:
{
"mcpServers": {
"globalwebmcp": {
"command": "node",
"args": ["/absolute/path/to/globalwebmcp/dist/src/index.js"]
}
}
}Teach it a site
The repository includes a project-local Codex skill. Run Codex from this checkout, then ask:
$map-site-action map article search on WikipediaThe skill drives the browser itself. It checks for native WebMCP, a stable URL or form, accessible DOM, and relevant same-origin requests. It then writes the smallest binding that preserves the observed result.
When a replay fails, run the skill again. It reproduces the failure, repairs the binding, bumps the map version, and reruns the checks. Runtime execution does not guess when a site changes.
How it fits together
website
│
│ browser discovery
▼
map-site-action skill
│
│ writes a reviewed contract
▼
maps/<site>.json
│
│ strict startup validation
▼
local Global WebMCP server
│
├── find_sites
├── list_site_actions
├── prepare_site_action
└── execute_site_actionThe number of MCP tools stays fixed as the catalog grows. Sites add data, not another server process or another tool namespace. Each map owns one site and a set of versioned actions. See the complete maps/wikipedia-en.json example.
What it will and will not run
Binding | Catalog support | Execution |
Read-only public HTTPS URL | Yes | Yes |
OAuth API | Yes | Blocked with required provider and scopes |
Browser session | Yes | Blocked with the prepared same-origin URL |
Native WebMCP | Yes | Blocked with the required tool name |
Write, transaction, or destructive action | Yes | Blocked |
The current executor is intentionally strict:
The map fixes the URL. Callers cannot supply or override it.
JSON Schema validates every input before rendering.
Public actions use
GETon the declared HTTPS origin.The server does not follow redirects.
Requests time out after 15 seconds.
Responses stop at 1 MiB.
Map text and website output remain untrusted.
The map format has no fields for credentials, cookies, tokens, or browser state.
This is the useful first cut. Browser-session execution and authenticated writes belong in later adapters because they need session ownership, approval, and stronger completion checks.
Top-site options
These options follow Similarweb's worldwide ranking for July 2026, updated August 1, 2026. Each site was inspected in a live browser before its map was written.
Rank | Site | Versioned action | Current execution |
1 |
| Browser session required | |
2 | YouTube |
| Browser session required |
3 |
| Browser session required | |
4 |
| Browser session required | |
5 | ChatGPT |
| Write and browser session blocked |
6 | X |
| Browser session required |
7 |
| Browser session required; CAPTCHA observed | |
8 | Bing |
| Executable public |
9 | TikTok |
| Executable public |
10 |
| Write and browser session blocked |
The catalog also includes executable Wikipedia and GitHub searches, a blocked GitHub issue action, and a blocked Amazon UK product search.
Travel search options
These actions preserve the full search intent without entering a property or reservation flow.
Site | Versioned action | Inputs | Current execution |
Booking.com |
| Destination, check-in, checkout, adults, and rooms | Browser session required |
Airbnb |
| Destination, check-in, checkout, and adults | Browser session required |
Develop
npm run typecheck
npm test
npm run maps:check
npm run verify:wikipediaThe test suite validates the protocol, catalog, strict input boundary, blocker policy, and MCP stdio flow. The Wikipedia check calls the real local MCP server and the live Wikipedia API.
globalwebmcp/
├── .agents/skills/ # autonomous website mapper
├── maps/ # versioned site contracts
├── src/protocol.ts # map schema and safety rules
├── src/catalog.ts # immutable action index and planner
├── src/executor.ts # bounded public HTTP execution
├── src/server.ts # four MCP tools over stdio
├── scripts/ # map and live verification
└── tests/ # protocol, catalog, and MCP checksThe design record lives in docs/design/0001-v0-catalog.md.
License
MIT. Build the browser bridge only when a real site forces you to.
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 Connectors
Create, edit, preview, publish, and manage web pages from MCP-capable AI clients.
Agent-first web hosting: deploy sites, apps, databases and domains over MCP.
Hosted MCP for creating, checking, deploying, and hosting static sites for AI agents.
AI agent website builder. Create and publish link-in-bio sites via MCP or REST API.
Related MCP Servers
- FlicenseNot gradedqualityAmaintenanceEnables MCP-compatible AI agents to execute typed live-app actions over WebSocket, allowing agents to directly interact with real application state without browser automation or scraping.20
- FlicenseNot gradedqualityAmaintenanceEnables external AI agents to edit and manage a self-hosted website via a token-gated MCP server, including page content, data, media, and sandboxed functions with validation and rollback.3
- FlicenseNot gradedqualityCmaintenanceEnables AI agents to interact with any website through MCP, providing structured knowledge graph contexts, generated actions, and readiness scoring.
- AlicenseNot gradedqualityBmaintenanceEnables AI agents to draft, review, preview, and publish content across multiple websites through typed MCP tools. Ensures safe, policy-gated, and auditable publication without requiring a conventional CMS admin interface.Apache 2.0
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/benjang032/globalwebmcp'
If you have feedback or need assistance with the MCP directory API, please join our Discord server