Skip to main content

Plugins

Siesta AI plugins connect another AI assistant to your Siesta AI account. They let you ask the assistant to work with accessible Siesta AI agents and data from the host application. The connection uses the Model Context Protocol (MCP) and your Siesta AI permissions.

The plugin is an integration for an AI assistant. It is separate from the Browser Extension, which adds Siesta AI features to your browser.

Architecture​

The host loads integration metadata and connects to the remote MCP server at https://api.siesta.ai/mcp. The host manages OAuth sign-in and presents the user with the available tools; Siesta AI authenticates requests, filters the tool catalog by permissions, and remains the system of record for agents, conversations, and other resources.

User → ChatGPT / Codex / Microsoft 365 Copilot
     → OAuth-authenticated MCP connection → Siesta AI API → accessible resources

The production MCP scope is mcp; offline_access can enable refresh tokens when supported by the client. Availability, schemas, and actions come from the connected server: a plugin package is not a guarantee that every tool is available to every account. Consult MCP setup for connection and authorization details.

How it works​

  1. Open the Siesta AI integration in the host application, if it is available in your environment.
  2. Connect your Siesta AI account through the application's sign-in flow.
  3. Ask the assistant to find an agent, inspect a conversation, or work with other accessible Siesta AI resources. Available actions depend on the integration, your organization, and your permissions.
  4. Review the assistant's response and the outcome of any requested action. If a run is still pending or needs approval, follow the instructions shown by the integration and continue in Siesta AI when necessary.

The host application sends requests to the Siesta AI MCP server on your behalf. Connecting an account does not grant access to resources that account cannot use. Keep passwords and access tokens out of chat messages.

ChatGPT and Codex​

The Siesta AI plugin package for ChatGPT and Codex uses an authenticated connection to https://api.siesta.ai/mcp. It includes instructions for discovering accessible agents, starting or continuing their conversations, and inspecting results. Depending on the server catalog and account permissions, other tools for skills, Memory, Graphs, and workflows may also be available.

The package contains .codex-plugin/plugin.json for plugin metadata, .mcp.json for the remote server and OAuth declaration, and a Siesta AI skill that guides tool use. Credentials are managed by the host's connection flow. The package uses ordinary remote MCP tools; it does not itself add an embedded Siesta AI panel.

For a conversation request, the assistant can discover an agent with list_agents, use send_message with an agentId for a new conversation or a conversationId to continue one, and inspect history with get_conversation. A tool response can indicate a finished, pending, failed, or approval-required state. Check the returned state before describing the outcome; pending work is not a completed answer. Do not repeat a message after an uncertain timeout without checking the conversation.

For example, ask: “List my Siesta AI agents and explain what they do.” To use an agent, identify it and describe the request. A pending conversation is not a completed result; continue in Siesta AI if the integration cannot complete the action. For approval-required actions whose full arguments cannot be verified through the plugin, review the action in Siesta AI.

The repository contains a submission package. Its presence alone does not confirm public directory availability or successful authentication in every ChatGPT or Codex environment. Follow your administrator's installation instructions for the version available to you.

Microsoft 365 Copilot​

The Microsoft 365 Copilot integration is a separate package prepared for Microsoft Partner Center. It connects Microsoft 365 Copilot to the same Siesta AI MCP endpoint using OAuth. The available tools are discovered dynamically and depend on the signed-in Siesta AI account.

The package combines a Teams app manifest.json, a declarative agent definition, and ai-plugin.json. The plugin definition declares a RemoteMCPServer runtime with OAuthPluginVault and dynamic discovery (run_for_functions: ["*"]). The Microsoft OAuth registration is configured on the Microsoft side; its reference is not an API key and does not replace Siesta AI authorization. Dynamic discovery does not bypass the user's Siesta AI permissions.

You can ask for accessible agents, conversations, skills, Memory, Graphs, or workflows. For example: “Which Siesta AI agents can I use?” or “Show my workflows and their recent runs.” The package instructs the assistant to describe proposed changes and wait for confirmation before actions such as sending a message or running a workflow.

You need a Siesta AI account and access to Microsoft 365 Copilot in your Microsoft environment. An administrator must make the integration available. The checked-in package does not by itself establish marketplace approval or a successful live Microsoft sign-in and tool call.

For connection details, see MCP setup. If an integration or resource is unavailable, ask your administrator to check its deployment and your access.

Last updated on