Cekura

npx add-mcp 'https://api.cekura.ai/mcp'

Set up Cekura in your AI assistant

Install Cekura Skills and MCP so your AI assistant can design, run, and improve voice-agent evaluations.

[!NOTE]

Recommended setup: install both Cekura Skills and the Cekura MCP server.

Skills teach your assistant how to work with Cekura. MCP lets it safely act in your Cekura workspace.

Set up

Install the Cekura plugin for your assistant — it bundles Skills and MCP. Pick your tab.

Claude Code

The plugin bundles Skills, slash commands, and MCP config in one install. In a Claude Code session, run:

/plugin marketplace add cekura-ai/cekura-skills
/plugin install cekura@cekura-skills

The plugin already includes the MCP server config; /setup-mcp signs you in (OAuth) and verifies the connection. Once the plugin loads (restart the session if prompted), run:

/setup-mcp

OAuth opens a browser to sign in — no API key. Then run /cekura-onboarding for a guided first run. Full guide →

Claude Desktop

The plugin bundles Skills + MCP (Skills work in Chat and Cowork). In Claude Desktop:

1. Open Customize → Plugins
2. Personal plugins → + → Add marketplace → Add from a repository, and enter cekura-ai/cekura-skills
3. Install cekura, and authorize MCP via OAuth when prompted

Full guide →

Cursor

The plugin includes Skills + MCP. In Cursor:

1. Open Settings → Plugins → Team Marketplaces → Add Marketplace → Import from Repo
2. Enter https://github.com/cekura-ai/cekura-skills
3. Install cekura, and authorize MCP via OAuth when prompted

Full guide →

Codex

The plugin bundles Skills + MCP. Add the marketplace, install, then authenticate:

codex plugin marketplace add cekura-ai/cekura-skills
codex plugin add cekura@cekura
codex mcp login cekura

codex mcp login opens a browser for OAuth — no API key. Full guide →

GitHub Copilot

The plugin bundles Skills + MCP for Copilot CLI. Add the marketplace, then install:

copilot plugin marketplace add cekura-ai/cekura-skills
copilot plugin install cekura@cekura-skills

OAuth runs in the browser on first use of a Cekura tool — no API key. Full guide →

Gemini CLI

Install the extension — it includes the Cekura MCP server and context:

gemini extensions install https://github.com/cekura-ai/cekura-skills

OAuth runs on first use of a Cekura tool. Details →

Which setup fits you?

Setup	Best for	What you get
⭐ Native plugin	Claude Code, Claude Desktop, Cursor, Codex, Copilot CLI, Gemini CLI	Skills + MCP, auto-configured (Claude Code also adds slash commands + hooks)
Manual (Skills + MCP)	Clients without a plugin	Install Skills, then connect MCP separately
Skills only	Any Agent Skills client, no live tools	Workflow guidance via npx skills add; no MCP
Behavior preset	Windsurf & assistants without a plugin	Domain knowledge via rules / AGENTS.md; no MCP

Per-client install commands are in Set up above. Not sure? Use the plugin — it's the least setup.

Verify your setup

After setup, ask your assistant:

Use the Cekura skills and MCP. List my Cekura agents, pick one, and propose 3 evaluators I should create first. Do not create anything until I approve.

You should see two things:

• The assistant can access your Cekura workspace and list real agents.
• The assistant reasons in Cekura terms like evaluators, metrics, expected outcomes, test profiles, and run results.

[!NOTE]

On an MCP-only setup (no Skills), verify with just List my Cekura agents — you'll get live tool access but not the workflow guidance that Skills add.

Why install both?

Skills add judgment

Skills tell your assistant what good Cekura work looks like: how to design evaluators, choose metrics, debug failing runs, and improve prompts safely.

MCP adds action

MCP gives your assistant permissioned tools to list agents, create metrics, generate evaluators, run tests, inspect transcripts, and analyze results.

With Skills only, your assistant can advise. With MCP only, your assistant can call tools. With both, it can choose the right workflow and complete it.

What you can ask next

First-time onboarding

I'm new to Cekura. Walk me through setting up my first agent, metrics, evaluators, and test run.

Create evaluator coverage

Read my agent's purpose and propose a coverage plan with workflow, edge-case, and regression evaluators. Wait for approval before creating them.

Run scenarios over the right connection

Each scenarios_run_* tool drives one connection type, and only works if that connection is configured on the agent. Common ones include scenarios_run_voice (telephony), scenarios_run_sip (SIP), scenarios_run_vapi_webrtc / scenarios_run_retell_webrtc (provider WebRTC), and scenarios_run_text (chat) — with more for ElevenLabs, LiveKit, Pipecat, Agora, and raw WebSocket. Retrieve the agent first and check provider.type and the telephony block to pick the right one.

Design better metrics

Create a metric that checks whether the agent resolves the caller's issue without escalating unnecessarily. Show me the rubric before saving.

Triage failed tests

Pull the failures from my latest Cekura run, group them by root cause, and tell me which ones are real bugs versus flaky runs.

Reference

Skills Catalog

See every Cekura Skill, install path, update path, and compatibility note.

Manual setup (Skills + MCP)

For clients without a plugin: install Skills, then OAuth/API-key MCP config for Claude Code, Claude Desktop, Cursor, VS Code, Codex, and Copilot CLI.

This documentation is built and hosted on Mintlify, a developer documentation platform.