Owlex
Get a second opinion in Claude Code: query Codex, Gemini, OpenCode and more, with structured multi-model deliberation.
Add to your client
Copy the config for your MCP client and paste it into its config file.
uv tool install git+https://github.com/agentic-mcp-tools/owlex.gitPaste into ~/Library/Application Support/Claude/claude_desktop_config.json
{
"mcpServers": {
"owlex": {
"command": "owlex-server",
"args": []
}
}
}Step-by-step guides: Add to Claude Desktop · Add to Cursor · Add to Windsurf
Before you start
- Python 3.10+
- uv (for installation via `uv tool install`)
- External agent CLIs as needed (Codex, Gemini, OpenCode, AiChat)
- Existing subscriptions for Codex/Gemini, or API tokens for OpenCode/AiChat
- OPENROUTER_API_KEY to enable the ClaudeOR agent
About Owlex
Owlex is a Python MCP server that brings multiple external AI agents into Claude Code for second opinions and structured deliberation. Instead of relying on a single model, you can run a multi-round council where Codex, Gemini, OpenCode, ClaudeOR, and AiChat answer independently, review each other's responses, and let Claude synthesize a final answer. It also supports specialist roles, team presets, individual agent sessions, and async background task management.
Tools & capabilities (15)
council_askRun a structured council deliberation: agents answer independently, optionally revise or critique each other, and Claude synthesizes a final answer. Supports claude_opinion, deliberate, critique, roles, team, and timeout options.
start_codex_sessionStart a new Codex session.
resume_codex_sessionResume a Codex session with a session ID or `--last`.
start_gemini_sessionStart a new Gemini session.
resume_gemini_sessionResume a Gemini session with an index or `latest`.
start_opencode_sessionStart a new OpenCode session.
resume_opencode_sessionResume an OpenCode session with a session ID or `--continue`.
start_claudeor_sessionStart a new Claude-via-OpenRouter (ClaudeOR) session.
resume_claudeor_sessionResume a ClaudeOR session with a session ID or `--continue`.
start_aichat_sessionStart a new AiChat session.
resume_aichat_sessionResume an AiChat session with a session name.
wait_for_taskBlock until a background task completes.
get_task_resultCheck a task result without blocking.
list_tasksList tasks with an optional status filter.
cancel_taskKill a running task.
When to use it
- Getting a second opinion on architecture decisions (e.g., monorepo vs. multiple repos)
- Debugging tricky issues by comparing multiple models' reasoning
- Running a security audit or code review with specialist agent roles
- Increasing confidence on high-stakes decisions via multi-model deliberation
- Querying large codebases with Gemini's 1M context window from within Claude Code
Security notes
Owlex orchestrates external AI CLIs/agents that can run with elevated permissions. `CODEX_BYPASS_APPROVALS=true` bypasses the Codex sandbox and `GEMINI_YOLO_MODE=true` auto-approves Gemini actions - use these with caution. `OPENCODE_AGENT` defaults to `plan` (read-only); set to `build` only when you want write access. Agents like ClaudeOR and AiChat require an `OPENROUTER_API_KEY` and incur API token costs.
Owlex FAQ
How does the council deliberation work?
In Round 1 each agent answers your question independently without seeing others. In Round 2 each agent sees all Round 1 answers and can revise its position (or critique others, if critique mode is on). Finally Claude reviews everything and outputs a structured synthesized answer.
Which agents are supported?
Codex, Gemini, OpenCode, ClaudeOR (Claude via OpenRouter), and AiChat. You can exclude agents with the COUNCIL_EXCLUDE_AGENTS environment variable to control cost.
What are specialist roles and team presets?
Roles (security, perf, skeptic, architect, maintainer, dx, testing, neutral) shape how an agent analyzes a problem. You can assign them explicitly per-agent, auto-assign from a list, or use predefined team presets like security_audit, code_review, architecture_review, devil_advocate, balanced, and optimal.
Does it cost money to run?
Codex and Gemini use your existing subscriptions, while OpenCode and AiChat use API tokens (provider-dependent). The maintainers recommend using the council for important decisions rather than every question, and excluding agents to manage cost.
Alternatives to Owlex
Compare all alternatives →
Official MCP server providing persistent, file-backed knowledge-graph memory across sessions.
Structured step-by-step reasoning tool for breaking problems into revisable thought sequences.
Fully managed remote server for AWS docs, blogs, What's-New and Well-Architected guidance — no key.