MCP Quickstart

Five minutes, one key, nothing to install. The server is remote, and it authenticates with a single API key.

Step 1 — Create an API key

1. Sign in at app.conbersa.ai.
2. Open Settings → API keys.
3. Create a key and copy it. It looks like cb_live_... and is shown once.

Full details in API keys.

Step 2 — Add the Conbersa server to your client

Use this endpoint and header:

URL:    https://api.conbersa.ai/mcp
Header: Authorization: Bearer cb_live_YOUR_KEY

The exact config depends on the client:

• ChatGPT — custom connector with API-key auth.
• Claude — Claude Code / Desktop with a bearer header.
• Cursor — mcp.json server entry.
• opencode — remote MCP entry in opencode.json.

Step 3 — Verify with a read-only call

Ask the agent:

List my Conbersa devices.

It calls list_devices and returns your fleet with pad_id, platform, username, and a busy flag. If you see your devices, both auth and access work.

Step 4 — Run a safe first task

Pick one free device and a short task:

Start a run on pad <pad_id> that likes two videos and stops.

The agent calls start_run, which returns a group_id and a session link. Open the link to watch it work. Then stop it:

Stop that run.

That calls stop_run. You've now used the whole loop: read, run, status, stop.

Step 5 — Graduate to saved prompts

Free-form tasks are fine for tests. For work you'll repeat, use saved prompts:

List my prompts, then run the one called "Warmup — TikTok" on these three devices.

Saved prompts carry persona, brand, model, and step limits, so runs stay consistent. See MCP workflows for patterns like fleet-wide posting and scheduled distribution.

Troubleshooting a failed connect

More in MCP troubleshooting.