> This page is for version v1.0.1 (default).
> For other versions, use one of these documentation indexes:
> - v1.0.1 (default): https://docs.kinetk.ai/v-1-0-1/llms.txt

> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.kinetk.ai/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.kinetk.ai/_mcp/server.

# MCP Overview

> Use the KINETK graph from Claude, ChatGPT, Cursor and other AI assistants through the kinetk MCP server.

The **`kinetk` MCP server** is a [Model Context Protocol](https://modelcontextprotocol.io) server that exposes the KINETK Knowledge Graph as tools inside your AI assistant. Mention **kinetk** in a prompt and the assistant browses trending narratives, walks the visual pattern map and runs live retrieval jobs for you, getting back token-efficient answers.

## Two ways to run it

|            | **Hosted MCP** (recommended)                                      | **Local MCP** (advanced, self-run)                                   |
| ---------- | ----------------------------------------------------------------- | -------------------------------------------------------------------- |
| What it is | A managed endpoint KINETK runs for you                            | The `kinetk-mcp` package you run as a subprocess                     |
| Setup      | Nothing to install — sign in, or pass an `x-api-key` header       | Clone, build, and point your client at it                            |
| Tools      | The full set: context jobs, narratives, patterns, search/fetch    | The three original context-job tools only                            |
| Best for   | Claude Desktop, ChatGPT, Claude Code, Codex, Cursor, Gemini, Grok | Pinning a specific build, or custom forks                            |
| Guide      | [Hosted MCP Setup](/documentation/mcp-server/hosted-mcp-setup)    | [Local MCP (Advanced)](/documentation/mcp-server/local-mcp-advanced) |

Most users want the **[Hosted MCP](/documentation/mcp-server/hosted-mcp-setup)**: it is always current and has every tool. Both bill the same account at the same prices as the [HTTP API](/documentation/get-started/pricing-credits).

## What you can do with it

Ask your assistant what's trending, dig into a narrative, run a full analysis, explore what content looks like, or pull posts to cite. See [What You Can Ask](/documentation/mcp-server/what-you-can-ask) for example prompts and what each costs.

## When to use the MCP vs the API

| Use case                                                                           | Pick                                    |
| ---------------------------------------------------------------------------------- | --------------------------------------- |
| You work in an AI assistant and want it to fetch live context                      | **MCP**                                 |
| You're exploring a topic conversationally — trending, drill-down, related patterns | **MCP**                                 |
| You're building a backend, scheduled job or dashboard                              | **API**                                 |
| You need `POST /search`, `stats` jobs, `GET /usage` or `GET /health`               | **API** — these aren't exposed as tools |

## Why slim results?

Full results can run to hundreds of kilobytes, and assistants pay per token. The job tools return a slim, LLM-optimized envelope by default; ask for `verbose: true` on `kinetk_get_context_job_result` when you need the full payload.