This is the single-page setup for Claude Desktop. Everything you need — Cekura skills, the MCP server, and your own custom skills — installs in a few minutes through the Claude Desktop plugin marketplace.Documentation Index
Fetch the complete documentation index at: https://docs.cekura.ai/llms.txt
Use this file to discover all available pages before exploring further.
One-line summary: install the
cekura plugin from https://github.com/cekura-ai/cekura-skills.git via Settings → Plugins, finish the MCP OAuth sign-in, and you’re done. Custom skills live in the ~/.claude/skills/ folder or in your own marketplace fork.What you’ll set up
- Cekura plugin — bundles the nine Cekura skills and auto-configures the MCP server.
- MCP authorization — one-click OAuth into your Cekura workspace.
- Custom skills — your own organization-specific playbooks, installed alongside the Cekura ones.
Prerequisites
- A Cekura dashboard account.
- A current Claude Desktop build (Mac or Windows) with Plugins enabled. On Team/Enterprise, a workspace admin must enable Plugins and Connectors at the org level first.
- No Node.js, no API keys, and no manual config-file editing for the default OAuth path.
1. Install the Cekura plugin
The plugin ships all nine Cekura skills plus the MCP server configuration in a single install — samecekura-ai/cekura-skills marketplace used by Claude Code.
Add the Cekura marketplace
Click Marketplaces → Add marketplace. Paste:Confirm. Claude Desktop fetches the marketplace manifest.
Install the `cekura` plugin
Switch to the Discover (or Plugins) tab, search
cekura, and install. This registers the nine skills and adds the Cekura MCP server to your connector list.Authorize the MCP
Open a new chat. On the first Cekura tool call, Claude Desktop opens a browser for OAuth — sign into your Cekura dashboard and click Authorize. The session token is managed for you.
Verify
Ask: “Use the Cekura MCP to list my agents, then use the Cekura skills to propose 10 workflow evaluators covering the most common flows for the first agent.”Claude should call
aiagents_list, load the cekura-eval-design skill, and draft evaluators. If you get 401 Unauthorized, reconnect from Settings → Connectors → Cekura.What you get out of the box
| Capability | What it does |
|---|---|
| 9 Cekura skills | Auto-activate on phrases like “design a metric for…” or “fix this failing run…” — full catalog in Skills overview. |
| Cekura MCP server | 84+ tools for agents, evaluators, metrics, runs, and call logs. Configuration in MCP overview. |
| Skill activation logs | Visible in Claude’s chain-of-thought when the matching skill loads. |
Slash commands (
/create-metric, /run-evals, etc.) are a Claude Code-only feature today. In Claude Desktop, you describe the task in plain English and Claude auto-activates the matching skill. The end result is the same — Claude calls the same MCP tools under the hood.2. Add your own custom skills
Custom skills encode organization-specific knowledge — naming conventions, compliance checks, internal runbooks — and activate the same way the Cekura ones do.Skill structure
A skill is a folder with one required file,SKILL.md, plus any supporting files it references.
SKILL.md starts with YAML frontmatter — name and a specific description (Claude matches your message against the description to decide whether to load the skill).
Ways to install
| Path | When to use |
|---|---|
Drop into ~/.claude/skills/<name>/ | Solo iteration. Restart Claude Desktop to pick it up. |
| Fork the marketplace, install the plugin | Team rollout — same Settings → Plugins → Add marketplace flow as Cekura, pointing at your fork. |
| Workspace upload | Team/Enterprise only — admin adds the marketplace at the org level. |
skill-creator shortcut, and the full authoring guide, see Anthropic’s skills documentation and the Agent Skills standard.
3. End-to-end example
A typical “set up evals for a new agent” loop in Claude Desktop:- “List my Cekura agents.” — MCP returns agents.
- “For agent
support-bot, generate 10 workflow evaluators following our naming conventions.” — auto-loadscekura-eval-designand your custommy-team-eval-conventionsskill, drafts the evaluators, calls the MCP to create them. - “Run every evaluator on
support-botand stream pass/fail.” — MCP runs them concurrently. - “For the 3 failures, rerun each 4 times and tell me which are real bugs vs. flakes.” — Claude triages with
cekura-fixing-prod-issues. - “Draft a fix plan with the suspected prompt or tool changes for each real bug.” — Claude composes the report inline.
Troubleshooting
| Symptom | Fix |
|---|---|
| Plugin install fails or marketplace not found | Confirm the URL is exactly https://github.com/cekura-ai/cekura-skills.git. Re-run Marketplace → Refresh. |
401 Unauthorized from MCP tools | Settings → Connectors → Cekura → Reconnect to re-run OAuth. |
| Cekura tools don’t appear in a chat | Restart Claude Desktop after installing the plugin. Confirm the Cekura connector is Enabled. |
| Skills don’t activate | Confirm Settings → Capabilities → Skills is on, restart the app, and check that the skill’s description: matches the phrasing you’re using. |
Custom skill in ~/.claude/skills/ isn’t loading | The folder must contain SKILL.md at its root (not nested), and the name: frontmatter field must match the folder name. |
| Team workspace doesn’t show Plugins or Connectors | An admin needs to enable them once at the org level. |
References
MCP Overview
Full configuration matrix, auth options, request headers
Skills Catalog
All nine Cekura skills and what each one activates on
Claude Code Guide
The Claude Code equivalent (plugin marketplace + slash commands)
Dashboard
Manage projects, agents, and API keys