Using Openclaw as MCP with Codex, Claude Code, Claude ai and regular Openclaw agents

[lament wave]

I often have a problem with Claude Code, Claude.ai and Codex not  I think that the important aspects of OpenClaw and some are wondering if I can have OpenClaw so happen on this in PCP so they can get access to the documentation and the underpinnings of OpenClaw. As well as keep running mugs at 10 plus OpenClaw agents at the same time.

[lone parrot]

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

[deep mango]

I think you mean MCP — and yes, mostly.

Use the paths like this:

1. Claude Code / Codex / MCP clients talk to OpenClaw

   • Use openclaw mcp serve
   • This exposes OpenClaw conversations/history/reply tools over MCP.
   • Docs: https://docs.openclaw.ai/cli/mcp

2. OpenClaw runs Claude Code / Codex / Gemini as agents

   • Use ACP agents: /acp doctor, then /acp spawn claude ...
   • For Codex specifically, prefer native /codex ... unless you explicitly need ACP.
   • Docs: https://docs.openclaw.ai/tools/acp-agents

3. 10+ OpenClaw agents

   • Use multi-agent routing for separate personas/workspaces/accounts.
   • Use sub-agents for parallel background work.

• Default sub-agent concurrency is 8; you can raise agents.defaults.subagents.maxConcurrent, but watch RAM, rate limits, and token cost.
  • Docs: https://docs.openclaw.ai/concepts/multi-agent and https://docs.openclaw.ai/tools/subagents

Small caveat: openclaw mcp serve exposes OpenClaw sessions/conversations, not automatically the whole OpenClaw source/docs. For docs/source access, point Claude/Codex at the repo/docs directly, or add a separate docs/filesystem MCP server.