Read your robot's manifest from OpenAI Codex CLI via the standard MCP stdio transport. One config snippet — same robot-md-mcp server as every other surface.
--dangerously-bypass
to enable MCP tool execution. Only use this in trusted environments.
Step 1 — Install the CLI
$ pip install robot-md
Step 2 — Launch Codex with the MCP server
$ codex --mcp-server "npx --yes robot-md-mcp ./ROBOT.md" \ --dangerously-bypass
# From your robot project directory: $ codex --mcp-server "npx --yes robot-md-mcp ./ROBOT.md" \ --dangerously-bypass > what can this robot do? Reading robot-md://my-robot/capabilities... This robot exposes: arm.pick, arm.place, arm.reach, vision.describe, status.report. Max payload 0.5 kg. Software E-stop at 100 ms.
| Feature | Codex CLI (MCP) | Notes |
|---|---|---|
| Read capabilities / safety / frontmatter | ✓ Native | MCP resources; requires --dangerously-bypass |
| Validate ROBOT.md | ✓ Native | Requires sandbox bypass |
| Render canonical YAML | ✓ Native | Requires sandbox bypass |
| Compliance filing commands | → CLI only | Run robot-md emit-fria from shell |
| Sandboxed mode (no bypass) | — Blocked | Codex sandbox blocks MCP tool calls; resource reads may also be blocked |