# New Features

## MCP Trigger

Frends Processes can now be exposed as tools that AI assistants and other Model Context Protocol clients can discover and invoke directly.

When an MCP Trigger is added to a Process, the Agent advertises that Process as a callable tool through an MCP server endpoint. External MCP clients — including Frends' own AI Connector and third-party agents such as Claude or Copilot — can list the available tools per Agent Group, inspect their JSON Schema, and execute them with structured inputs. The Trigger uses the official `ModelContextProtocol.AspNetCore` library, supports per-Process tool descriptions and input validation, and integrates with the existing Process logging.

Learn more about the [MCP Trigger](/reference/triggers/mcp-trigger.md).

## AI Connector — MCP Tool Discovery and Tool Calling

The AI Connector shape can now discover and call MCP tools published by Frends Agents and by external MCP servers.

A new Tools tab on the AI Connector configuration lists tools available in the selected Agent Group's Processes and lets you connect to external MCP servers — either by fetching their tool list immediately and selecting which tools are allowed for the connector to use or by providing providing the allowed tool names manually, the tools are actual tools resolved at runtime on the Agent and compared to the allowed list. Tools advertised by other Frends Processes can be selected and invoked as part of an AI Connector call, allowing AI shapes to orchestrate other Frends Processes through the MCP protocol. Reasoning steps are now opt-in to keep token usage and latency predictable for callers that don't need them.

Read more about the improved capabilities of [AI Connector with MCP tools](/reference/shapes/activity-shapes/ai-connector.md).

## RabbitMQ Trigger with Client Certificate Authentication

The RabbitMQ Trigger gained full support for mutual TLS authentication using client certificates.

In addition to the existing username/password authentication, the Trigger configuration now accepts a client certificate (PEM or PFX) and an optional passphrase, along with TriggerInput parameters for referencing certificates from the certificate store. This enables brokers that require client certificate authentication.

Check here tolearn more about [RabbitMQ Trigger with client certificates](/reference/triggers/mcp-trigger.md).

## Agent Registration Mode

The Agent Group configuration page replaces the boolean "Allow new agents to connect" toggle with an explicit Agent Registration Mode selector with four options: Self-Hosted Agent, Cloud Agent, Kubernetes Deployment, and Kubernetes StatefulSet.

Self-Hosted mode requires Agents to be created and removed manually through the UI. Kubernetes Deployment mode automatically registers Agents on heartbeat and removes them when they shut down, matching the previous "Allow Auto Connect" behaviour. Kubernetes StatefulSet mode registers Agents on heartbeat but never deregisters them automatically — fitting workloads with stable identities. The "Download Kube Config" action now produces the appropriate Deployment or StatefulSet manifest based on the selected mode, and mode changes are locked when active Agents are present in Self-Hosted or Cloud groups.

More details about [Agent Registration Modes from here](/guides/setup-and-installation/creating-new-environments-and-agent-groups.md).

## Remote Subprocess Execution Scalability

Remote subprocess execution has been re-architected with a less session intensive implementation. Instead of using per-execution sessions, the sessions are Agent specific and are internally routed to the correct Process execution.

A new Agent-level subprocess reply listener is auto-detected per Agent and per Agent Group, and the platform now batches and routes replies efficiently even under high parallel-subprocess loads. Customers using fan-out subprocess patterns will see significantly higher throughput and lower latency.

You can learn more about [remote Subprocesses here](/reference/process-development/subprocess.md).

## Improved mTLS (client certificate support)

Client certificate validation is now enabled by default for the Cross-platform Agent for HTTP and API Triggers — matching the Windows Agent's behaviour out of the box.

The OpenAPI support has been updated to 3.1.1 (which is the new default) which also supports mTLS authentication options.


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://docs.frends.com/release-notes/frends-6.3/new-features.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.
