Set up Skills + MCP manually

Most clients — Claude Code, Cursor, Codex, Gemini CLI — have a native plugin that installs Skills and MCP in one step. Use this page only if your client has no plugin. You’ll set up the two pieces separately: Skills first, then MCP.

What MCP gives you

Cekura’s MCP server exposes documentation search and Cekura API operations through the US endpoint by default:

https://api.cekura.ai/mcp

Regional endpoints

Use the endpoint that matches your Cekura workspace region. Keep the US endpoint above as the default unless your workspace is hosted in India or EU.

Workspace region	MCP endpoint
US (default)	`https://api.cekura.ai/mcp`
India	`https://api.in.cekura.ai/mcp`
EU	`https://api.eu.cekura.ai/mcp`

For India or EU, replace the US URL in the client examples below with the matching regional endpoint. Your assistant can search docs, list agents, create metrics, generate evaluators, run tests, inspect results, and analyze call logs through typed tools.

Prerequisites

Access to the Cekura Dashboard
Node.js 20.18.1 or higher if your client uses the mcp-remote shim
No Node.js requirement for Claude Code’s native HTTP transport, Claude Desktop connectors, or Codex native HTTP config

Install Skills

Install Cekura Skills first, so your assistant knows Cekura workflows — not just raw tools:

npx skills add cekura-ai/cekura-skills --all

If your assistant doesn’t load Agent Skills, use the portable behavior preset instead. (Only need raw tools, e.g. for CI? Skip ahead to Connect MCP.)

Connect MCP

Then connect the MCP server. OAuth is recommended; use an API key only for CI or project-scoped credentials.

OAuth setup

Your MCP client opens a browser, you sign into Cekura, and the client manages the session token.

Claude Code
Claude Desktop
Cursor / VS Code
Codex
Copilot CLI

claude mcp add --transport http cekura --scope user https://api.cekura.ai/mcp

Verify:

claude mcp list

Open Claude Desktop.
Go to Settings > Connectors.
Click Add custom connector.
Set Name to Cekura.
Set URL to https://api.cekura.ai/mcp.
Save, then finish the browser authorization flow.

This connector gives MCP only. To get Cekura Skills too, install the plugin instead — see the Claude Desktop guide.

Add this to your Cursor MCP config or .vscode/mcp.json:

{
  "mcpServers": {
    "cekura": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://api.cekura.ai/mcp"]
    }
  }
}

Add this to ~/.codex/config.toml. The mcp-remote shim handles the OAuth browser sign-in (needs Node.js 20.18.1+):

[mcp_servers.cekura]
command = "npx"
args = ["-y", "mcp-remote", "https://api.cekura.ai/mcp"]

The example uses the US default. For a regional Codex connection, use one of these commands instead:

# India
codex mcp add cekura-india --url https://api.in.cekura.ai/mcp
codex mcp login cekura-india

# EU
codex mcp add cekura-eu --url https://api.eu.cekura.ai/mcp
codex mcp login cekura-eu

Add this to ~/.copilot/mcp-config.json — Copilot handles the OAuth browser sign-in on first tool use:

{
  "mcpServers": {
    "cekura": {
      "type": "http",
      "url": "https://api.cekura.ai/mcp",
      "tools": ["*"]
    }
  }
}

Or add it from the command line:

copilot mcp add --transport http cekura https://api.cekura.ai/mcp

API-key setup

Use an API key only when you need a project-scoped credential, a shared service credential, or a client that cannot complete OAuth. Create keys in Settings > API Keys in the Cekura dashboard.

Claude Code
Claude Desktop
Cursor / VS Code
Codex
Copilot CLI

claude mcp add --transport http cekura --scope user https://api.cekura.ai/mcp --header "X-CEKURA-API-KEY:YOUR_API_KEY_HERE"

Verify:

claude mcp list

{
  "mcpServers": {
    "cekura": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://api.cekura.ai/mcp",
        "--header",
        "X-CEKURA-API-KEY:${CEKURA_API_KEY}"
      ],
      "env": {
        "CEKURA_API_KEY": "YOUR_API_KEY_HERE"
      }
    }
  }
}

Native HTTP config:

[mcp_servers.cekura]
url = "https://api.cekura.ai/mcp"
http_headers = { "X-CEKURA-API-KEY" = "YOUR_API_KEY_HERE" }

CLI form:

codex mcp add cekura --env CEKURA_API_KEY=YOUR_API_KEY_HERE -- sh -c 'npx -y mcp-remote https://api.cekura.ai/mcp --header "X-CEKURA-API-KEY:$CEKURA_API_KEY"'

Add this to ~/.copilot/mcp-config.json:

{
  "mcpServers": {
    "cekura": {
      "type": "http",
      "url": "https://api.cekura.ai/mcp",
      "headers": {
        "X-CEKURA-API-KEY": "YOUR_API_KEY_HERE"
      },
      "tools": ["*"]
    }
  }
}

CLI form:

copilot mcp add --transport http --header "X-CEKURA-API-KEY: YOUR_API_KEY_HERE" cekura https://api.cekura.ai/mcp

This is the credential the Copilot coding agent and Copilot code review need — those surfaces don’t support OAuth remote MCP.

Verify the connection

Restart your assistant and ask:

List my Cekura agents.

If configured correctly, your assistant returns agents from your Cekura workspace.

Request headers

Header	Used by	Description
`Authorization: Bearer <token>`	OAuth	Set automatically by your MCP client after sign-in
`X-CEKURA-API-KEY`	API-key path	Required when using an API key
`Content-Type: application/json`	Both	Request payload format

Available tool areas

Documentation Search

Search integration guides, schemas, examples, and workflow docs.

Testing

Create agents, create evaluators, define metrics, run tests, and inspect results.

Observability

List call logs, evaluate production calls, and analyze call performance.

Automation

Create scheduled jobs, recurring tests, and metric-improvement workflows.

Troubleshooting

Auth failed

A 401 Unauthorized usually means the OAuth session expired or the API key is invalid.For OAuth, reconnect from your client and confirm you are signed into the right Cekura account. For API keys, confirm the header name is exactly X-CEKURA-API-KEY and the key is active.

Tools are not showing up

Restart your assistant after changing MCP config. Confirm the URL matches your workspace region in the regional endpoint table, check JSON or TOML syntax, and inspect your assistant’s MCP logs.

Node.js version errors

If mcp-remote fails with an engine or File is not defined error, upgrade Node.js to 20.18.1 or higher and make sure your client is using that Node.js version.

Rate limits

If you hit 429 Too Many Requests, reduce concurrency, narrow the request, or check usage in Settings > API Usage.

Next step

If your client turns out to have a native plugin, switch to it — it bundles Skills + MCP and keeps both updated. See the setup overview. Browse the Skills catalog for what the Skills cover.

Get started

Set up your assistant

Reference

Workflows

Set up Skills + MCP manually

What MCP gives you

Regional endpoints

Prerequisites

Install Skills

Connect MCP

OAuth setup

API-key setup

Verify the connection

Request headers

Available tool areas

Documentation Search

Testing

Observability

Automation

Troubleshooting

Next step

​What MCP gives you

​Regional endpoints

​Prerequisites

​Install Skills

​Connect MCP

​OAuth setup

​API-key setup

​Verify the connection

​Request headers

​Available tool areas

Documentation Search

Testing

Observability

Automation

​Troubleshooting

​Next step

What MCP gives you

Regional endpoints

Prerequisites

Install Skills

Connect MCP

OAuth setup

API-key setup

Verify the connection

Request headers

Available tool areas

Troubleshooting

Next step