Orq Claude Plugins

Orq plugins for Claude — tracing, MCP tools, and agent skills. Works with Claude Code, Claude Desktop, and Claude Cowork.

Claude Desktop (non-technical setup)

If you're using the Claude Desktop app (not the terminal), see the Desktop Setup Guide — no command line needed.

The short version: add this to your claude_desktop_config.json (see the guide for where to find it):

{
  "mcpServers": {
    "orq": {
      "type": "http",
      "url": "https://my.orq.ai/v2/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_ORQ_API_KEY_HERE"
      }
    }
  }
}

Claude Code Installation

# Add the marketplace
claude plugin marketplace add orq-ai/claude-plugins

# Install whichever plugins you need
claude plugin install orq-trace@orq-claude-plugin
claude plugin install orq-mcp@orq-claude-plugin
claude plugin install orq-skills@orq-claude-plugin

Configuration

API key

Both the tracing and MCP plugins require an orq.ai API key. Set ORQ_API_KEY in your shell, .env file, or as a command prefix:

ORQ_API_KEY=sk-... claude

Trace plugin variables

Variable	Description
ORQ_TRACE_PROFILE	Orqi profile for trace destination (highest priority — decouples traces from CLI/MCP)
ORQ_API_KEY	API key (used if no trace profile is set)
ORQ_BASE_URL	Orq API base URL (default: https://my.orq.ai)
ORQ_TRACE_USER	User identity attached to traces (falls back to git config user.email)
ORQ_TRACE_MAX_CONTENT_LEN	Truncate content exceeding this character limit
TRACE_ORQ_REDACT_CONTENT	Strip all input/output bodies from traces
OTEL_EXPORTER_OTLP_ENDPOINT	Override the OTLP endpoint directly
OTEL_EXPORTER_OTLP_HEADERS	Extra headers for the OTLP endpoint
ORQ_DEBUG	Enable debug logging to stderr (true/1)

Tracing auto-enables when an API key can be resolved (via ORQ_TRACE_PROFILE, ORQ_API_KEY, or orqi CLI profile).

See plugins/trace-hooks/README.md for the full configuration reference.

Plugins

orq-trace

Automatically trace Claude Code sessions to Orq. Captures sessions, turns, tool calls, and LLM responses as hierarchical OTLP spans sent to /v2/otel/v1/traces.

Span tree:

orq.claude_code.session
├── claude_code.turn.1
│   ├── chat claude-opus-4-6
│   ├── execute_tool Read
│   ├── chat claude-opus-4-6
│   ├── execute_tool Edit
│   └── subagent.Explore
├── claude_code.turn.2
│   └── ...

orq-mcp

Orq MCP server — gives Claude Code access to Orq platform tools via https://my.orq.ai/v2/mcp.

orq-skills (submodule → orq-ai/orq-skills)

Agent skills for the Build → Evaluate → Optimize lifecycle on orq.ai. Includes:

• Skills — build-agent, build-evaluator, run-experiment, optimize-prompt, analyze-trace-failures, generate-synthetic-dataset, prompt-learning
• Commands — /orq:quickstart, /orq:invoke, /orq:workspace, /orq:traces, /orq:analytics, /orq:models

See the orq-skills README for full documentation.

Development

Test plugins locally:

claude --plugin-dir ./plugins/trace-hooks
claude --plugin-dir ./plugins/mcp
claude --plugin-dir ./plugins/skills

Updating

After making changes, bump the version in both marketplace.json and the relevant plugin.json, then push. Users update with:

claude plugin marketplace update orq-claude-plugin

To update the skills submodule to the latest upstream:

git submodule update --remote plugins/skills

License

MIT