docs / search-docs

Search the Kronos docs from your AI

Let Claude Desktop, ChatGPT, n8n, or any AI client search this documentation in natural language. Available through MCP and as an open REST endpoint.

What this feature is

Kronos ships a search tool that any AI client can call to look up the Kronos documentation in natural language. Instead of pasting docs into your prompt, your AI asks Kronos and gets back the relevant sections.

There are two ways to reach it:

Through your MCP client

Connect Kronos as an MCP server in Claude Desktop, ChatGPT, n8n, or any MCP-compatible app. A tool named search_docs is automatically available alongside the messaging tools. Your AI decides when to call it. See /docs/mcp for the one-time setup.

As an open REST endpoint

A simple HTTPS POST that anyone can call without authentication — handy for code, integrations, or one-off scripts.

Not the same as your Kronos AI agents

The AI agents you configure inside Kronos (the ones you assign to a WhatsApp / IG / Messenger inbox) answer your customers using your own knowledge base — the PDFs, docs, audio, and product info you uploaded. They do NOT search the Kronos product docs while talking to your customers, and there is no setting that turns that on.

search_docs is a separate, opt-in surface for AI clients outside Kronos (Claude Desktop, ChatGPT, n8n, your own scripts) that want to query the Kronos product docs. Connecting your personal Claude Desktop to Kronos's MCP server gives it access to this tool — it never bleeds into your customer-facing agents.

If you ever want a customer-facing agent to know specific bits of Kronos product info, the right place to put them is your knowledge base — not this search tool.

What you can ask

Anything you'd normally search the docs for, in your own words:

·"How do I connect my WhatsApp Business number?"
·"How much does Kronos AI cost in MXN?"
·"How do follow-ups work in agentic mode?"
·"Can my agent read images that customers send?"
·"How do I send a template message through the API?"

The tool reads the same docs you're looking at right now, returns the most relevant sections, and tells your AI which doc each result came from so it can cite back.

Using it from your AI client

Once Kronos is connected as an MCP server (one-time setup, see /docs/mcp), just ask your AI in plain language:

"Search the Kronos docs: how do I enable file understanding?"

Behind the scenes the AI calls search_docs with your question, Kronos returns the matching sections, and the AI answers using that content. No code on your side.

Using it from code (open REST)

For developers or non-MCP integrations, the same search is available as a public POST. No API key required.

curl -X POST https://api.kronos.com.mx/v1/docs/search \
  -H "Content-Type: application/json" \
  -d '{
    "query": "how do I send a WhatsApp template message",
    "limit": 3
  }'

The response is a JSON object with an items array — each item carries the matching doc, the section heading, the text content, and a relevance score from 0 to 1 (higher = closer match).

Which surface should I use? (token vs no token)

Both surfaces run the exact same search. They differ only in authentication — pick by how you connect:

REST — no token. POST /v1/docs/search is fully open. Right for an anonymous agent, a script, or a quick lookup.
MCP — needs an API key. The MCP server also exposes private account tools (sending messages, reading conversations), so the whole server is gated behind a Bearer key and search_docs shares that gate. No key? Use the REST endpoint.

Two more machine-readable surfaces: every docs page has a markdown twin at /docs/<slug>.md, and a curated index for LLMs lives at /llms.txt.

Arguments

Argument	Required	Notes
query	yes	The natural-language question. 1–500 characters.
limit	no	How many results to return. 1–20, defaults to 5.
doc_slug	no	Restrict the search to one specific doc (e.g. `whatsapp`, `api`, `mobile-app`). Slug format is lowercase with hyphens.

Limits

The open REST endpoint is rate-limited to 20 requests/minute/IP, plus a global daily ceiling across all callers (an anti-abuse cap that normal usage never reaches). Queries longer than 500 characters are rejected, and repeated identical queries are served from a short-lived cache. The MCP version is rate-limited per API key like every other Kronos tool — see /docs/mcp.