plori Codex plugin

A one-install plugin for plori: persistent cloud environments for AI agents, with durable disks, real tools, and memory that stays with each agent.

Installing this plugin gives Codex two things at once:

• The plori MCP server (.mcp.json): the remote server at https://api.plori.ai/mcp, so create_agent, invoke_agent, schedule_run, and the rest become tools in your Codex session. Authentication is OAuth 2.1, auto-detected on first use (sign in once with an email code), or an API key.
• The plori skill (skills/plori/SKILL.md): teaches Codex how to connect, create agents, invoke them and read replies, answer human-in-the-loop requests, and schedule deferred runs. Its content is the same one served at https://plori.ai/.well-known/agent-skills/plori/SKILL.md.

Install

Add this marketplace, then install the plugin:

codex plugin marketplace add plori-ai/codex-plugin
codex plugin add plori@plori

The plori server is authenticated over OAuth 2.1 the first time Codex reaches a tool; approve the one-time email sign-in when the browser opens. Running an agent spends credits on your plori account.

Or connect the MCP server directly (no plugin)

The plugin is a convenience wrapper. You can also add the remote server on its own:

codex mcp add plori --url https://api.plori.ai/mcp
codex mcp login plori

Background runs

A plori run often takes longer than the tool call that started it. Codex gives an MCP tool 60 seconds by default, and the plori server returns within that: invoke_agent holds your call for about 50 seconds, then answers with a run_id and a poll_after_seconds delay. Call get_run_result again after that delay until the run completes or reports awaiting_input.

For work that runs for minutes, use the plori CLI, version 0.4.0 or later:

curl -fsSL https://plori.ai/install.sh | sh
plori login
plori watch --events terminal,input &   # one JSON line per run that ends or pauses
plori inbox                             # the same information, one shot

Each plori watch line names the run. Read the reply with plori result <agent> <run-id>, and answer a paused run with plori answer <run-id> <tool-call-id> --approve.

To let a single tool call run longer instead, add the plori server to ~/.codex/config.toml yourself, as in Or connect the MCP server directly, and set tool_timeout_sec on that entry:

[mcp_servers.plori]
url = "https://api.plori.ai/mcp"
tool_timeout_sec = 900

This plugin does not ship that value. The documented fields for a plugin-bundled remote MCP server are type, url, and headers; none of them set a timeout. The plugin-scoped config keys (plugins."plori".mcp_servers.plori.*) cover enabling a server, its tool allowlists, and tool approval, not timeouts.

What is inside

.codex-plugin/plugin.json           # the plugin manifest
.mcp.json                           # remote MCP server (api.plori.ai/mcp)
skills/plori/SKILL.md               # the plori skill
.agents/plugins/marketplace.json    # self-marketplace (lists this plugin at ./)
SECURITY.md                         # vulnerability disclosure policy

Nothing here runs local code or installs third-party software: the plugin only points Codex at the hosted plori MCP server and adds the skill text.

Links

• Site: https://plori.ai
• MCP connect guide: https://plori.ai/mcp
• Integration front door: https://plori.ai/agents.md
• CLI on npm: https://www.npmjs.com/package/@plori/cli
• Questions: dev@plori.ai

License

MIT