mastra-ai/mastra @mastra/core@0.16.2

2025-09-10

on GitHub

Agents

• A bundling issue by marking 'typescript' as an external dependency in both '@mastra/agent-builder' and '@mastra/server', preventing unwanted TypeScript chunks in the build output. #7585

CI / Tests

• Fixes the peerdeps action to ensure proper functionality and compatibility. #7535
• Add a missing changeset for #7566 to ensure the Braintrust version is incremented in the next release. #7596
• Comprehensive tests for the convertFullStreamChunkToMastra function, increasing coverage by 116 lines and validating its handling of various stream chunk types and edge cases. #7637

CLI / Playground

• The CLI development startup copy, changing references from 'Local' to 'Playground'. #7579
• The Storybook configuration for the Playground UI. #7584

Client SDK - JS

• An issue in AI SDK V5 where Gemini models incorrectly handled image inputs by ensuring that image URLs and base64 data are properly distinguished during message conversion, preventing decoding errors. #7490
• The execution of clientTools in the JavaScript client SDK when using agent streamVNext. #7577
• A structuredOutput property to client-js, making the model parameter optional and defaulting to the agent model if not provided. #7597

Core Platform Components

• A bug by removing a duplicate 'float' case from a switch statement. #7516
• The passing of the usage property to streamVNext's onFinish and adds usage and total usage tracking. #7598
• The documentation by adding Mastra class methods to the navigation. #7601

Deployer

• The schedule to run the Discord-to-GitHub triage process daily via a cron job. #7505
• The Vercel deployer to support HTTP PATCH requests by adding the PATCH method to its entry file. #7492
• A bug in @mastra/deployer by correcting the usage of flags with yarn pack to properly handle native dependencies in yarn monorepos. #7570
• Bugs with the transpilePackages feature during mastra dev by adding stricter checks and changing the transpilation output location for workspace packages to improve reliability and avoid embedding third-party dependencies. #7572
• Missing peer dependency errors by explicitly adding '@opentelemetry/api' to the output 'package.json' during deployment. #7518
• The AI exporter by setting a default cloud endpoint, clarifying the environment variable name, renaming the exporter for consistency, and updating the README. #7600

Developer Tools & UI

• Support for cursor slash commands. #7528
• Playground UI components to enhance Observability features. #7496
• An unintended body scrollbar by adding overflow-hidden to the body, preventing unnecessary scrolling caused by Radix Popover components. #7631

MCP

• Rollback a previous change that added the createMCPTool helper for proper execute types in the MCP module. #7513
• That MCP tools are recreated and properly populated with the mastra instance when attaching the MCP server, fixing initialization timing issues. #7520
• Fix the peer dependency for MCP package. #7533
• Output schema injection for the vNext version, enhancing how output schemas are handled. #6990
• An issue where external image URLs in message file parts were incorrectly converted to base64 data URIs, ensuring URLs are now handled and validated properly without unwanted transformation. #7542
• The documentation for runtimeContext usage in agents and tools. #7583
• Consistent messageId values in responses by instantiating a single MessageList per call in both stream/generate VNext and legacy paths, resolving issues with mismatched IDs between responses and the database. #7606
• The server API route types to make them available for external use. #7657

Memory

• That getMemoryMessages is called during the first turn in a thread when semanticMemory is scoped to resource, addressing issue #7367. #7529
• The agent memory tests to improve or modify their coverage or behavior. #7536

Networks

• The documentation for Agent.streamVNext. #7639

Observability

• Comprehensive, parameterized integration tests for AI tracing in various mastra agent configurations, featuring intelligent mocking, advanced debugging, and coverage of core scenarios to validate functionality and highlight future implementation needs. #7515
• A new Observability playground page. #6752
• The Observability Trace Dialog by changing the 'Span Id' label to '#' for consistency and fixing the heading style. #7618

Prod analytics

• 'projectName' as a top-level field to the Braintrust exporter and updates usage metrics to the BraintrustUsageMetrics format for improved trace reporting. #7566

Storage

• A new @mastra/s3vectors package, providing a first-class vector store for Mastra backed by Amazon S3 Vectors (Preview), supporting index management, vector CRUD operations, strict metadata filtering, and integration with the official AWS SDK, along with comprehensive tests and documentation. #6801
• The listIndexes() method in @mastra/pg to only return tables managed by Mastra, preventing conflicts with external pgvector tables and adds tests to ensure correct handling of such cases. #7539

Tools

• Concurrency to the tool execution step for stream/generate vnext, improving performance by enabling parallel execution. #7524
• Made the TracingContext parameter optional in tool.execute(), allowing manual calls without requiring it. #7532
• A critical bug in tracing context propagation for tool calls, ensuring proper span hierarchy and observability in nested agent, tool, and workflow executions, and updates tests for complex scenarios. #7531
• Support for provider-executed tools by checking a flag to skip local execution and return pre-computed results, ensuring correct handling of tools like OpenAI's webSearchPreview and Claude's str_replace_editor, with accompanying tests. #7642

Website

• The documentation to fix broken links. #7603
• The sidebar to modify the observability link. #7621
• A previous change that updated the sidebar observability link. #7635

Workflows

• An issue where the workflow span appeared in a separate trace, ensuring correct trace linking in the core workflow logic. #7522
• The playground to use the new workflow streamVNext feature when streaming. #7575
• Comprehensive tests for the Workflow.createRecordStream method in the JavaScript client SDK, increasing coverage by 15 lines and verifying both success and error scenarios for streaming record serialization. #7594
• AI tracing propagation for workflows executed directly within agents. #7599
• Comprehensive tests for the getStep utility, improving coverage of workflow navigation and nested EventedWorkflow resolution logic. #7433
• Dynamic workflow execution within the agent thread as a new feature. #7506
• Workflow documentation by moving workflow methods into a dedicated methods directory and adds a redirect. #7617
• The codebase to use createRunAsync, likely refactoring existing functionality to support asynchronous run creation. #7632
• Options are correctly passed to streamVNext, flattens LoopConfig callbacks into execution options, adds tests, and updates documentation. #7643
• An issue where the wrong repository slug was used to fetch template environment variables, ensuring the correct slug from the URL is used to prevent 404 errors. #7650
• The prepareStep option to the generateVNext and streamVNext functions in the core module. #7646
• The Workflow type and adds support for workflow-step-output. #7519