STACK · SEP 27, 2026 · 5 MIN
How to Connect Cruva or Euka to Claude to Automate Creator Management
Step-by-step setup for the Cruva MCP and the Euka MCP in claude.ai and Claude Code, the prompts worth running first, and the guardrails we use before letting Claude act on creators.
To connect Cruva or Euka to Claude, add the vendor's MCP server as a custom connector: https://mcp.cruva.com for Cruva, https://app.euka.ai/api/mcp for Euka. In claude.ai that is Settings, then Connectors, then Add custom connector; in Claude Code it is one claude mcp add command. After you sign in, Claude can read your creator data and, if you let it, take actions. Clankers builds these setups for TikTok Shop brands, and we start every one read-only.
Key takeaways:
- Both Cruva and Euka publish an MCP endpoint that signs in with OAuth. No code needed for claude.ai.
- Cruva says its MCP is included on every plan. Euka says MCP needs its Growth, Business, or Autopilot plan.
- Connected, Claude can send DMs and approve or reject samples. Treat that as a permission you grant later.
- Start with read-only prompts, keep a human on every creator-facing action, and never auto-approve samples.
How do I connect Cruva to Claude?
Cruva's help center documents the claude.ai steps (Cruva help center: Cruva MCP and API):
- Go to claude.ai/customize/connectors, or in the Claude app open Settings, then Connectors. Click Add custom connector.
- Name it Cruva and paste
https://mcp.cruva.com. Click Add. The same URL also appears in Cruva under Dashboard, then MCP. - Cruva now shows in your connectors list. Click Connect and sign in.
- Check you are signed in to the correct Cruva account, then click Allow access. For agencies, this account decides which shops Claude can see.
- Open a chat and ask. You can toggle Cruva on or off per chat from the tools menu.
Cruva says this takes about two minutes and that the MCP and APIs "are included on every plan."
How do I use the Cruva MCP with Claude Code?
Claude Code adds remote MCP servers with one command (Claude Code docs: MCP):
claude mcp add --transport http cruva https://mcp.cruva.com
Then run /mcp inside Claude Code and complete the Cruva sign-in in your browser. The add command saves the config without checking credentials, so sign-in happens at the /mcp step.
One trap Cruva calls out: do not send your Cruva REST API key as a Bearer token. That returns an error like "expected JWE compact serialization." Either let the client run the OAuth sign-in, or use the API-key URL Cruva documents, https://mcp.cruva.com/sse?api_key=<your_key> (Cruva help center).
Claude Code is the better home once you want the same job to run the same way every week, because the instructions live in a file your team can read and edit instead of a chat.
How do I connect Euka to Claude?
Euka's MCP page lists the endpoint https://app.euka.ai/api/mcp and four steps: add a custom MCP server in your AI client, paste the endpoint, sign in to Euka with OAuth, and start a chat (Euka MCP). In claude.ai that is the same Add custom connector flow as above. In Claude Code:
claude mcp add --transport http euka https://app.euka.ai/api/mcp
Then /mcp to sign in. Euka states MCP access requires its Growth, Business, or Autopilot plan.
What can I ask Claude once Cruva or Euka is connected?
Three kinds of request, in the order we introduce them.
Fetch data. Read-only, safe from day one.
How many samples did we ship last week, and how many of those creators have posted?
Which creators drove GMV for us in the last six months but have not posted in 45 days?
Produce work. Still read-only, but Claude assembles something a person used to build by hand.
Give me yesterday's recap: affiliate GMV, units, videos posted, top 3 creators, anything unusual vs the 7-day average.
That prompt is Cruva's own example. The version we install runs on a schedule and posts to Slack; see how to automate TikTok Shop weekly reporting. A scored version of the same data is the automated affiliate health score.
Take action. Claude drafts, a person approves, then it sends.
Draft a re-engagement DM to each creator on that list. Show me every draft before sending anything.
For how an MCP works in plain terms, read what is an MCP.
What guardrails should I set before Claude acts on creators?
Cruva's help center is direct about it: the MCP "can draft and send DMs and approve/reject samples, not just read data." That is useful and it is also the reason for rules. Ours:
- Read-only first. Run fetch and report prompts until the numbers match what you see in the dashboard. Only then add actions.
- Never auto-approve samples. The MCP can approve samples, so it is tempting to hand Claude a threshold rule. We do not. Every sample decision goes to a person, because a sample is product out the door.
- Draft, then approve. Any DM, invite, or email is shown in full before it sends. Ask for the drafts in a list you can scan.
- Name the shop. Agencies should state the shop in every prompt. Cruva says the signed-in account decides which shops Claude sees, and that agencies can pivot across all their shops in one conversation.
- Check the log. Cruva records API and MCP actions in the audit log, tagged with the LLM platform. Filter by platform weekly to see what Claude did.
- Write it down. Put the rules in the instructions Claude runs from, so they hold when a different person runs the job.
What goes wrong when connecting?
- The action failed. Cruva's first fix is to reconnect: re-run the sign-in at claude.ai/customize/connectors.
- Claude sees the wrong shops. You signed in with the wrong account. Disconnect and sign in again with the one that holds the shops.
- 401 or JWE error in a coding tool. You sent the REST key as a Bearer token. Use OAuth or the
sse?api_key=URL. - No push events. Cruva notes there is no public webhook today, so anything recurring has to run on a schedule rather than react instantly.
Should I build this myself or have it built?
If someone on your team is comfortable in Claude Code, the connection is the easy part. The work is deciding what runs, on what schedule, with what rules, and making sure it survives when that person leaves. Clankers builds the workflows and Claude automations inside your stack and trains your team to own them. Our AI Advisory coaches technical teams to do it themselves.
Want to know which Cruva workflow to automate first? Get the free audit.