For AI agents: a documentation index is available at the root level at /llms.txt and /llms-full.txt. Append /llms.txt to any URL for a page-level index, or .md for the markdown version of any page.
kinetk.ai
DocumentationAPI Reference
  • Get Started
    • Introduction
    • Authentication
    • Quickstart
  • Guides
    • Narrative Intelligence
    • Intelligence Jobs
    • Sync & Freshness
  • MCP Server
    • Overview
    • Installation
    • Tools
kinetk.ai
On this page
  • What you can do
  • When to use the API vs the MCP
  • Base URL
  • Get started
Get Started

Introduction

Was this page helpful?

Authentication

Next
Built with

The KINETK Graph Service API is the public surface of the KINETK Knowledge Graph — a multimodal content intelligence platform that retrieves, clusters, and explains social-media narratives across TikTok, Instagram, YouTube, Reddit, Pinterest, X, and more.

This site documents two ways to use it:

  • HTTP API — REST endpoints behind an API key. Best for backends, dashboards, and scheduled jobs.
  • MCP server — a Model Context Protocol server (kinetk-mcp) that wraps the heavy async endpoints as agent-friendly tools. Best for Claude, Cursor, Gemini CLI, and other AI coding/research agents.

Both share the same backend, the same auth model, and the same caching/freshness guarantees.

What you can do

CapabilityAPI pathAsync?
Liveness + sync-freshness probeGET /healthsync
Precomputed trending narrativesGET /narratives/trendingsync
Filtered narrative searchGET /narratives/searchsync
One narrative with content + creators + breakdownGET /narratives/{id}sync
One creator profile by idGET /creators/{id}sync
Batch creator lookup (up to 500 ids)GET /creators?ids=...sync
Heavy retrieval: intelligence_search, intelligence_discover, campaign_brief, llm_contextPOST /intelligence/jobs + GET /intelligence/jobs/{id}async

The async path solves the API Gateway 29-second ceiling: submit returns immediately with a jobId, then poll until status: succeeded. Identical inputs within a per-kind freshness window are served from cache. See Intelligence Jobs for the full lifecycle.

When to use the API vs the MCP

Use caseRecommended path
You’re building a backend service or dashboardHTTP API directly
You’re scripting a one-shot ETL or backfillHTTP API directly
You’re working in Claude / Cursor / Gemini CLI and want the AI to fetch contextMCP server
You’re prototyping an agent that orchestrates queriesMCP server

The MCP doesn’t expose every endpoint — it surfaces a curated set of three intent-shaped tools (create_context_job / get_context_job_status / get_context_job_result) that map to the async /intelligence/jobs flow. Direct HTTP gives you everything; MCP gives you the most useful subset with token-efficient response envelopes.

Base URL

The production API is served at:

https://api.kinetk.ai/graph

Every endpoint requires an x-api-key header. We recommend exporting both as env vars so the examples elsewhere in these docs work directly:

$export API_BASE="https://api.kinetk.ai/graph"
$export API_KEY="<your key>"

The MCP server uses the same values via GRAPH_SERVICE_URL and GRAPH_SERVICE_API_KEY — see MCP Installation.

Get started

  1. Authentication — get an API key.
  2. Quickstart — curl your first request.
  3. MCP installation — connect Claude/Cursor/Gemini to the server.