WebMCP vs org MCP — two products, one trust story
If you remember one diagram from HUMΛN’s MCP story, make it this:
| Org MCP | WebMCP trust layer | |
|---|---|---|
| Where | https://{org}.mcp.haio.run |
Browser window.webmcp + HUMΛN bridge |
| Who calls | Cursor, Claude Desktop, IDE agents | Pages and extensions on the open web |
| What you expose | Your workflows, agents, catalog | Site tools inbound/outbound under HumanOS |
| Threat focus | Delegation abuse inside the org mesh | Origin, approval, data minimization on arbitrary sites |
Same company. Same taste for receipts. Different products.
The hook: the wrong README
A team reads “MCP” once, wires the org endpoint into a browser experiment, and wonders why origin scoping and external adapters feel bolted on. Or they expect {org}.mcp.haio.run to wrap every website tool on earth. Both mistakes come from collapsing two nouns into one acronym.
Org MCP in one sentence
Your org as an MCP server — governed colleague to the platform you already run. See Your org is an MCP server.
WebMCP in one sentence
A browser trust layer for WebMCP tools: facade (expose HUMΛN capabilities) and adapter (discover/govern external site tools), both through identity, delegation, approval, capability routing, minimization, and provenance.
Transport essays (Streamable HTTP) explain how bytes move. This essay explains which product you meant.
So that buyers and builders pick deliberately
Say “org MCP” when you mean AI clients calling your mesh. Say “WebMCP” when you mean browser tool governance. If a deck mixes them, send them this table before the pilot starts.
Scroll-stopper: One word — MCP — two threat models. Pick the noun before you pick the URL.
Go deeper
- Product: Developers, HumanOS
- Docs: MCP integration (org mesh); WebMCP deep dive in monorepo
docs/deep-dives/webmcp-trust-layer.md - Community: Org mesh, Streamable HTTP, Passport/delegation
MCP power surfaces — Part 5 of 6
Code & Docs