Mastra just turned Claude Code, Cursor, and Codex into subagents you orchestrate

Mastra shipped SDK subagents on June 15.

You can now run Claude Code, Cursor, and OpenAI Codex as subagents inside a single Mastra workflow — with full stream composition, evals, and Studio observability.

This is not a wrapper. It is an agent meta-harness.

Each coding agent keeps its native feature set — file handling, codebase reasoning, multi-agent handoffs, permission controls — while inheriting Mastra's agent surface: .generate(), .stream(), .resumeGenerate(), .resumeStream().

The architectural implication is real.

Your Mastra orchestrator can delegate a refactoring task to Claude Code, a multi-file edit to Cursor, and a test generation run to Codex — then compose their outputs through a workflow step or subagent delegation.

Swap one harness for another without changing your orchestration code.

Install: npm install @mastra/claude @anthropic-ai/claude-agent-sdk

Requires @mastra/core@1.38.0 or later.

The coding agent landscape just went from "pick one" to "compose all three."

If your team is still debating Claude Code vs Cursor vs Codex, stop choosing. Orchestrate.

Audit your agent harness stack. Map which coding agent handles which task type. Wire them into a single Mastra workflow with shared observability. You get harness redundancy without harness lock-in.

SOURCE: https://mastra.ai/blog/introducing-sdk-subagents
VERIFIED: Mastra blog (June 15, 2026), GitHub PRs #16906 and #17525, npm package @mastra/claude
SIGNAL: Mastra is positioning itself as the orchestration layer above coding agents, not a competitor to them. This is the "Kubernetes for coding agents" play — own the scheduler, not the workload.
---