NLWeb vs WebMCP: when to use each.
Match your need to the right protocol. Read top-to-bottom or jump to the row that sounds like your problem.
| Need | Best fit | Why |
|---|---|---|
| I want my product catalog to be answerable by AI search | NLWeb | Server-side semantic indexing of structured content is exactly NLWeb's job. |
| I want an agent to help users complete a multi-step booking on my site | WebMCP | The flow needs live page state and the user's session — only a client-side agent can drive it. |
| I want my SaaS dashboard to be operable by an agent the user brought | WebMCP | Admin actions behind auth run in the authenticated browser session, not headless. |
| I want my agent to delegate research to another specialist agent | A2A | Agent-to-agent delegation is a different category from exposing a website to an agent. |
| I want both: indexable content AND interactive tools | NLWeb WebMCP | Run NLWeb for discovery and WebMCP for in-session actions. See “Build both.” |
| I want one of my agents to be discoverable by other agents | A2A | Publish an A2A agent card with skills so other agents can find and delegate to it. |
| I'm building an internal tool and want my own LLM client to operate it | WebMCP | WebMCP if it's browser-based; talk MCP directly if it isn't. |
I want my product catalog to be answerable by AI search
Server-side semantic indexing of structured content is exactly NLWeb's job.
I want an agent to help users complete a multi-step booking on my site
The flow needs live page state and the user's session — only a client-side agent can drive it.
I want my SaaS dashboard to be operable by an agent the user brought
Admin actions behind auth run in the authenticated browser session, not headless.
I want my agent to delegate research to another specialist agent
Agent-to-agent delegation is a different category from exposing a website to an agent.
I want both: indexable content AND interactive tools
Run NLWeb for discovery and WebMCP for in-session actions. See “Build both.”
I want one of my agents to be discoverable by other agents
Publish an A2A agent card with skills so other agents can find and delegate to it.
I'm building an internal tool and want my own LLM client to operate it
WebMCP if it's browser-based; talk MCP directly if it isn't.
The gray zones
Sometimes the answer is “it depends.” A content site with a single interactive search box might reasonably ship NLWeb alone and skip WebMCP. A heavily authenticated app with little public content leans WebMCP-only, even though its data could in principle be indexed. When both discovery and in-session action matter, you want both — which is its own design exercise.
See: building a site that speaks both →