> ## Documentation Index
> Fetch the complete documentation index at: https://docs.orq.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Orq for coding agents

> Choose between Orq Skills, the Orq CLI, and the Orq MCP server to bring Orq.ai agents, evaluation, and traces into a coding agent.

Three surfaces bring **Orq.ai** into a coding agent. They share one workspace and one API key, and they compose: `orq setup` can wire all three at once.

| Surface | Scope |
| - | - |
| [**Orq Skills**](/ai-studio/integrations/code-assistants/orq-skills) | Multi-step workflows the agent runs from a description: build an agent, align a judge, analyze traces, run an experiment. Needs a shell, plus the **Orq MCP** server connected for the workspace calls its steps make. |
| [**Orq CLI**](/reference/cli) | The widest surface: the generated command for every API operation, plus `orq connect`, `orq setup`, `orq launch`, and `orq doctor`. Runs from a terminal, a Makefile, or CI. |
| [**Orq MCP**](/ai-studio/integrations/code-assistants/orq-mcp) | 38 typed tools the agent calls mid-conversation without a shell: agents, datasets, evaluators, experiments, models, traces, and docs search. |

## Which one to use

* **The agent can run a shell** (Claude Code, Codex, OpenCode, Kimi Code, Kilo, Pi): Skills plus the CLI. Skills carry the workflow; the CLI carries it out.
* **The agent cannot run a shell** (Claude Desktop, claude.ai, Linear agents): MCP.
* **CI, or anything that must run without an agent present**: the CLI, or the `evaluatorq` SDK for experiments.

The job-by-job split is in [MCP tools or the CLI?](/ai-studio/integrations/code-assistants/orq-mcp#mcp-tools-or-the-cli).

## Supported coding agents

`orq setup` detects the agents on the machine and wires the ones it finds. The CLI supports eight: Claude Code, Codex, OpenCode, Kimi Code, Kilo, Pi, GitHub Copilot CLI, and Gemini CLI.

| Step | Command | What it does |
| - | - | - |
| Set up | `orq setup` | Sign in, create a project-scoped API key, and wire the detected agents |
| Wire permanently | `orq connect [agent] [capability]` | Write `gateway`, `tracing`, `skills`, or `mcp` into each agent's own config |
| One session only | `orq launch <agent>` | Start the agent routed through the **AI Gateway**, leaving nothing on disk |
| Diagnose | `orq doctor` | Check auth, config, and endpoint reachability |

See [Coding agents](/reference/cli#coding-agents) for the flags, including `--local`, `--dry-run`, and per-agent model overrides.

## Next steps

<CardGroup cols={3}>
  <Card title="Orq Skills" icon="wand-magic-sparkles" href="/ai-studio/integrations/code-assistants/orq-skills">
    Workflows the agent runs from a description.
  </Card>

  <Card title="Orq CLI" icon="terminal" href="/reference/cli">
    Install and run the full command surface.
  </Card>

  <Card title="Orq MCP" icon="https://mintcdn.com/orqai/E6QxcuOkIZbPb-u-/images/logos/mcp.svg?fit=max&auto=format&n=E6QxcuOkIZbPb-u-&q=85&s=85ff775ba1532474fb9d6b4e81adc322" href="/ai-studio/integrations/code-assistants/orq-mcp" width="16" height="16" data-path="images/logos/mcp.svg">
    Typed tools for the agent session.
  </Card>
</CardGroup>


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.