MCP vs WebMCP vs /llms.txt
/llms.txt tells agents what to read, while MCP and WebMCP describe structured capabilities a product intentionally exposes. They solve different layers of agent readiness and work best when they point to the same public facts.
/llms.txt
A concise file at the site root that curates product, docs, policy, and action URLs for agents. It should be readable markdown, link to same-origin public pages, and avoid dumping every sitemap entry.
# /llms.txt # A concise map for AI agents. [products] - /products - /pricing [docs] - /docs - /docs/api [actions] - /contact-sales - /demo
WebMCP
WebMCP is a discovery layer for structured web capabilities such as docs search, pricing lookup, product search, or demo request flows. It helps agents find intentional actions instead of scraping buttons and guessing form behavior.
MCP
MCP describes tools and resources. A public website can advertise safe capabilities through a server card or manifest, while keeping authenticated or high-risk operations behind explicit authorization and product boundaries.
Agent Skills
Agent Skills manifests can describe repeatable public tasks and the documents or endpoints that support them. They are useful when a site wants agents to follow a consistent workflow instead of improvising from generic page text.
How they fit together
A practical agent-readiness stack is /llms.txt for curated reading, schema for entity clarity, and well-known protocol files for capabilities. Do not publish a protocol file that advertises actions your product does not actually support.
Do not confuse a reading map with an action protocol. /llms.txt can point to docs and policies, but it does not prove that an agent can safely perform a task.
- Do I need MCP, WebMCP, and /llms.txt together?
- No. Start with the layer that matches your product. Content-heavy sites may begin with /llms.txt, while products with real capabilities can add WebMCP or MCP discovery.
- Which one should I implement first?
- Most teams should start with readable public pages and /llms.txt, then add protocol discovery only when there are stable capabilities or workflows worth describing.
- Can /llms.txt advertise actions?
- It can link to action pages and explain context, but it is not an execution protocol. Use capability metadata when agents need structured action boundaries.