MCP gets the headlines, but discovery is the unsung hero. Here's the full RFC 9727 + AgentSkills + llms.txt stack unLocked CRM publishes so AI tools find us…
An MCP server is only useful if AI clients can find it. That's where the discovery stack comes in — a layered set of well-known files and conventions that let any agent locate, authenticate to, and use a service without manual configuration.
The Layers
1. RFC 9727 — /.well-known/api-catalog
A linkset pointing at the OpenAPI spec, docs, status page, terms, and privacy policy. The IETF standard for "what's at this domain that an automated client should know about."
2. AgentSkills v0.2 — /.well-known/agent-skills/index.json
The emerging standard for declaring what an AI agent can do at a domain. Includes service descriptors, structured data, sitemaps, RSS, and JWKS keys for Web Bot Auth.
3. llms.txt and llms-full.txt
Plain-text site overviews aimed at LLM crawlers. llms.txt is the index; llms-full.txt is the entire site condensed into a single markdown document agents can ingest in one fetch.
4. Markdown Twins
Every key marketing page (pricing, AI models, carriers, voice OS) ships an .md alternate served with Content-Type: text/markdown so AI clients prefer the clean text version over the rendered HTML.
5. Web Bot Auth (RFC 9421) JWKS
Public Ed25519 keys at /.well-known/http-message-signatures-directory let well-behaved bots sign their requests. We can rate-limit anonymous traffic without blocking legitimate AI crawlers.
Why This Matters
If a buyer's AI agent can't find your product, your product doesn't exist to them. Discovery is the new SEO.
Where This Fits In unLocked
- best CRM for insurance agents: How insurance CRMs compare on quoting, commissions, AI and price.