FPROTOCOLOPEN CRYPTO ALLOCATION INFRA
LLM ENTRY POINTS

LLM entry points

LLM-friendly entry points

We provide five standard discovery files spanning the major agent ecosystems, from the Anthropic llms.txt convention to the OpenAI plugin manifest. Below: what each file is for, and when frameworks find them automatically.

If you use Claude, ChatGPT, LangChain or LlamaIndex, in most cases you just hand the framework the root domain or the openapi.json URL and it auto-discovers the available tools by following these files.

openapi.jsonprimary
/openapi.json
MACHINE-READABLE SPEC

The authoritative source. Describes each endpoint's method, params, response schema and examples. LLM tool-calling frameworks consume it first.

llms.txtcontext
/llms.txt
CONCISE SITEMAP

Plain text, Markdown-structured, one plain-language line per endpoint plus a clickable curl URL. Good to drop into a system prompt for fast orientation.

llms-full.txtcontext
/llms-full.txt
FULL REFERENCE

Builds on llms.txt with a curl example, error model and cache semantics per endpoint; good for RAG retrieval or long-context models.

.well-known/ai-plugin.jsonmanifest
/.well-known/ai-plugin.json
OPENAI PLUGIN MANIFEST

The ChatGPT-plugin-era manifest; some agent runtimes still probe this path to discover the API.

.well-known/agents.jsonmanifest
/.well-known/agents.json
AGENT CAPABILITY MANIFEST

The newer agent manifest: declares auth, rate limits, transport and an OpenAPI pointer.

· — WHEN FRAMEWORKS AUTO-DISCOVER

When frameworks auto-discover

  1. 1

    Given a root domain

    the agent runtime probes /.well-known/ai-plugin.json and /.well-known/agents.json, then follows them to openapi.json.

  2. 2

    Given an openapi.json URL

    Claude tool use / OpenAI function calling can derive tool definitions directly, no extra config.

  3. 3

    Given llms.txt

    long-context models read it wholesale into the system prompt and immediately get a semantic map of every endpoint.