Choose your AI assistant.

Prefer your currently connected conversation. The command above is session-only. To remove its override, exit that session and start Codex without it.

If the original conversation was saved, the picker below can select it. Its syntax was checked; restoring that conversation with the connection override has not been verified. Confirm both original dates and actual tools. Ephemeral conversations may not be recoverable.

codex resume -c 'mcp_servers.auramatrix_candidate.url="https://auramatrix.ai/work/mcp"'

An optional saved connection uses user configuration. Check existing names first.

codex mcp add auramatrix_candidate --url https://auramatrix.ai/work/mcp

To remove that optional saved connection: codex mcp remove auramatrix_candidate. Confirm it is no longer loaded in a new session.

No automatic conversation restoration or hot reload is assumed. Do not place conversation IDs in shared links.

The expected directory endpoint is https://auramatrix.ai/work/mcp. Check tools inside the conversation you will continue. Keep any displayed error. Do not weaken permissions, reinstall, or substitute shell requests to claim native success.

Tool discovery and an actual native answer are separate checks. Open returned links to compare listing IDs, versions, source dates, costs and unknowns.

These are independent examples, not your current search or selected projects. Generic coding example: compare mini-SWE-agent and OpenHands.

search_catalog, get_catalog_entry, compare_catalog_entries, compare_deployment_options, browse_tasks, list_public_requests, get_public_request are public read-only tools. They do not execute third-party projects.

Ask naturally; say so if you only want Agents. Use the first candidates for a conditional shortlist, then request more when needed. Keep directory facts, outside research and the AI’s own judgment distinct.