MCP Server

Connect KINETK to Your AI Assistant

Your AI assistant is great at reasoning, but it doesn’t know what’s trending on TikTok this morning, which creators are driving a narrative, or how a hashtag is performing across platforms. KINETK does.

With one quick setup, you can connect KINETK to the AI assistant you already use (Claude, ChatGPT, Cursor, and others) and then just ask in plain English. Your assistant pulls live intelligence from the KINETK Knowledge Graph, does the heavy lifting behind the scenes, and answers you directly.

“What’s trending in the World Cup right now?” “Which creators are driving the luxury-watch conversation this month?” “Pull the top posts about sustainable sneakers so I can cite them.”

You ask. KINETK answers.

Ways to connect

Pick the method your assistant supports — they all reach the same KINETK Knowledge Graph and bill the same account.

MethodBest forHow you authenticate
Sign in (OAuth)Claude Desktop, ChatGPTClick Connect → sign in with your KINETK login. No key to copy.
API keyClaude Code, Cursor, Gemini CLI, scriptsPaste your KINETK x-api-key.
OAuth via your IdPGemini Enterprise (Agentspace)Register Gemini Enterprise as an OAuth client; admin setup. See Gemini Enterprise below.

⚠️ Claude Desktop’s “custom connector” only supports the sign-in method — it can’t send an API-key header. Use the sign-in URL for Desktop, and the API-key URL for Claude Code / Cursor.

Sign in — Claude Desktop / ChatGPT

  1. Open Settings → Connectors → Add custom connector.
  2. Paste the sign-in address: https://api.kinetk.ai/graph/connect
  3. Save, then click Connect / Sign in — a browser opens to the KINETK login. Sign in and approve.
  4. KINETK shows as connected, and the tools appear near the message box.

No key, no config files — just sign in.

API key — Claude Code / Cursor / Gemini CLI

Use the API-key address: https://api.kinetk.ai/graph/mcp with a header x-api-key: YOUR_KEY.

  • Claude Code (one line): 
    claude mcp add --transport http kinetk https://api.kinetk.ai/graph/mcp --header "x-api-key: YOUR_KEY"
  • Cursor: Settings → MCP / Connectors → Add → paste the address, add the x-api-key header, refresh.
  • Gemini CLI: add an HTTP server to ~/.gemini/settings.json (use the httpUrl + headers keys so the key rides along on every request): 
    1{
    2  "mcpServers": {
    3    "kinetk": {
    4      "httpUrl": "https://api.kinetk.ai/graph/mcp",
    5      "headers": {
    6        "x-api-key": "YOUR_KEY"
    7      }
    8    }
    9  }
    10}
    Start a new gemini session, then run /mcp list to confirm kinetk is connected and /tools to see its tools.

Need a key? Contact the KINETK team.

Note: the consumer Gemini app (gemini.google.com / mobile) can’t add custom MCP connectors yet — use the Gemini CLI above, or Gemini Enterprise below.

Gemini Enterprise (custom MCP)

Gemini Enterprise (Agentspace) connects to KINETK as a custom MCP data store. It uses OAuth 2.0 through your own identity provider — there’s no API-key option here — and only the StreamableHTTP transport (point it at the …/mcp address, not …/connect).

This is an admin-driven setup, not a one-click connect:

  1. Prerequisites. Have an Organization Policy Administrator enable custom MCP servers (it’s off by default), and grant your account the Discovery Engine Editor (roles/discoveryengine.editor) role.

  2. Register Gemini Enterprise as an OAuth client with your IdP (Okta, Azure AD, Google, …), setting the authorization redirect URL to https://vertexaisearch.cloud.google.com/oauth-redirect. Note the resulting Client ID, Client Secret, Authorization URL, Token URL and Scopes.

  3. Add the server. In the Google Cloud console → Gemini Enterprise → Data stores → Create data store, search Custom MCP Server (Preview), and choose Add MCP server.

  4. Fill the MCP Server Configuration:

    FieldValue
    MCP Server URLhttps://api.kinetk.ai/graph/mcp
    Authorization URL / Token URLfrom your OAuth registration (step 2)
    Client ID / Client Secretfrom your OAuth registration (step 2)
    Scopesspace-separated list from step 2
    MCP Server Descriptione.g. “KINETK social-media narrative intelligence”

  5. Save and complete the OAuth consent. The KINETK tools become available to your Gemini Enterprise agents.

The OAuth endpoints and client credentials are issued by KINETK — contact the KINETK team to provision them for your organization.

How to use it

Once connected, just talk to your assistant like always, and mention “kinetk” when you want live intelligence so it uses the connector instead of answering from memory.

“Use kinetk to tell me the latest trends in the World Cup.”

You’ll get synthesized takeaways — the narratives forming, which platform they’re concentrated on, which hashtags carry an engagement premium, and where the white space is.

Find the source content

“Use kinetk to pull the top posts about luxury watches this month.”

You’ll get the actual ranked posts and videos — platform, engagement, similarity, and enrichment (themes/tone) — so you can read, cite, or repurpose them.

Keep going

“Which of those creators is biggest on Instagram?” · “Turn the third insight into three post ideas.”

The assistant figures out what to look up, waits for results, and summarizes — you don’t manage any of it.

Tips for better answers

  • Say “kinetk.” Without the cue your assistant may answer from general knowledge and skip the live data.
  • Be specific. “Trends in women’s football on TikTok this week” beats “sports trends.”
  • Heavier questions take a moment. Quick content lookups return in seconds; deeper trend analysis can take up to a minute — your assistant waits and then answers.

About your account and usage

Whether you sign in or use an API key, usage counts against your KINETK account. Pulling content and intelligence draws on your plan; simply being connected does not. Keep your API key private — treat it like a password.

Troubleshooting

  • Assistant answers without using KINETK → add “use kinetk” and confirm the connector shows connected.
  • Claude Desktop “couldn’t connect to the server” → make sure you used the sign-in address (…/graph/connect), not the API-key address. Desktop can’t send a key header.
  • “Unauthorized” on the API-key method → check the address and that x-api-key has no stray spaces.
  • No Funds → add funds to your account.
  • A trend question is slow → deeper analysis runs a fuller pipeline; give it up to a minute.