Mastra Changelog 2025-03-20

This week we launched our very own MCP documentation server, shipped improvements to AgentNetwork, and continued to make enhancements to workflows.

Mastra MCP Documentation Server

As per user requests, we built a documentation server that allows an AI assistant or IDE with MCP support (like Cursor, Windsurf, or Cline) to query Mastra's complete knowledge base.

Here are a few sample prompts you might try using with our documentation server:

• "Add evals to my agent and write tests"
• "Does Mastra work with the AI SDK?"
• "I'm running into a bug with agent memory, have there been any related changes or bug fixes recently?"

We've seen that coding agents using this system are able to find the information they need to complete a task more quickly and efficiently.

To install, run pnpm create mastra@latest and select Cursor or Windsurf when prompted to install the MCP server. Further details on the feature can be found here.

New AgentNetwork primitive

The AgentNetwork class provides a way to create a network of specialized agents that can collaborate to solve complex tasks. Unlike Workflows, which require explicit control over execution paths, AgentNetwork uses an LLM-based router to dynamically determine which agent to call next.

In its latest implementation, the RoutingAgent has access to a single tool called "transmit.” This tool handles:

• Single agent calls
• Multiple parallel agent calls
• Context sharing between agents

AgentNetwork remains an experimental feature but is now available in Mastra’s alpha and latest versions.

Workflow improvements

The latest workflow updates focused on enhancing control flow and execution handling:

• We're also exposing event functionality in workflows now. A major addition is experimental .afterEvent() support. This provides more flexible workflow orchestration.

export const queryProcessingWorkflow = new Workflow({
    name: 'query-processing-workflow',
    // Define the schema for the initial trigger data
    triggerSchema: z.object({
        query: z.string().describe('The user query'),
    }),
    // Define events that can be triggered during workflow execution
    events: {
        userFeedback: {
            schema: z.object({
                feedback: z.string().describe('User feedback on the initial response'),
            }),
        },
    },
});


queryProcessingWorkflow
    .step(generateResponse)
    .afterEvent('userFeedback')
    .step(improveResponse)
    .commit();

• Suspend/resume functionality was improved, with suspend metadata now being stored in context.resumeData for better state management.
• Several bugs were fixed, including issues with the .after() method execution when handling multiple passes, proper execution of else branches when if-branches contain loops, and corrected handling of skipped conditions on awaited steps.
• Finally, the workflow visualization in playground UI was enhanced with resizable panels (note: we also improved agent visualization).

That’s all the major updates for this week! 🚀

Stay tuned for more…

… and if you’re impatient, here are 3 emojis as a hint for what’s coming next: 👤💬👤