Plugin
What this type is for
Choose Plugin when an application needs to give agents new actions, add automation steps, provide an embedded interface, or combine several flows. For example, an agent can create a payment link or find an order in an external CRM, while the embedded page lets the user connect the CRM, select an account, and manage the integration.
Capabilities you can enable
- HTTP tool builder — several actions with separate URLs, parameters, execution modes, and request signatures.
- MCP server — one MCP endpoint from which Senler loads available tools.
- Embedded page — an interface inside the cabinet where users configure and work with the installed plugin's features.
- Application actions — explicitly marked backend methods that AI can access through MCP to work with the application's page and prepare settings for its tools or steps.
- Configurable tool instance — a form that opens when an action is added to an agent and stores public and private settings for that particular connection.
- Automation steps — application actions that users add to an automation flow, configure, and connect to its next blocks.
Tools, automation steps, and the embedded page are configured independently. An application can use one of these capabilities or combine them.
Application actions are also enabled separately. They do not replace agent tools or automation steps: users do not add them to an agent or workflow manually. AI receives these methods through MCP only in projects where the plugin is installed and active.
How the application appears in a project
The user installs the application from the plugin catalog. Installation connects its capabilities to the project but does not add every action to every agent: the required tool is selected separately in the agent settings. If the application needs to read or change project data through the API, it also runs a regular OAuth authorization flow.
The user installation flow is described in Plugins.
Agent tools
In Agent tools, choose one connection method for the complete application: builder or MCP server. For an HTTP action, define its system name, clear Russian and English names and descriptions, endpoint, parameters, timeout, and execution mode. The handler must verify the request signature and use event_id as an idempotency key.
When a user needs to select an account or provide a secret while adding an action, enable a configurable instance and implement its form through Senler Bridge. Private data must not be returned by the public API or written to logs.
If AI should work with embedded-page data or prepare tool and step settings at the user's request, publish application actions as well.
Automation steps
In Automation steps, the application can publish its own actions for the automation editor. The developer defines input parameters, result fields, a webhook, and one continuation mode: a single output, fixed branches, or branches created by the embedded page while the user configures the step.
Automation steps are separate from agent tools: users add them to a process flow and can save their results to automation variables. For a configurable step, the same embedded page opens with the automation_step_configurator launch type.
Embedded page
The embedded page opens the plugin's user interface directly inside the cabinet. It can contain connection and account settings, data, reports, and other ways to work with the plugin. The application developer controls its exact contents.
The same page can support two flows: the installed plugin's general interface and a form for configuring one tool instance while adding it to an agent. Senler Bridge distinguishes them through context.launch.type: embedded_page and tool_configurator.
In the Embedded page, enter the primary HTTPS URL. A separate developer URL can point to a local or test environment for application team members. The page server verifies the one-time launch_code; senler_* query parameters help render the interface but do not prove access by themselves.
Through Senler Bridge, the page receives verified launch context and can support highlighting and permitted interactions with interface elements. Use stable data-ai-context-id values in the app.* namespace so the documentation and interface work consistently.
Embedding the page does not by itself grant the application access to project data or settings. To read or change them, use Senler AI OAuth and API with permissions approved by the user.
Before publication
Check installation in a test project, HTTP request signatures or MCP endpoint availability, timeout handling, embedded-page framing, OAuth permissions when used, and public documentation. Then submit the application for moderation from Catalog publication.