Anthropic ships Claude Managed Agents at Code w/ Claude, raises Claude Code API limits

Anthropic opened its annual developer conference, Code w/ Claude, in San Francisco on May 6 with a coordinated set of announcements aimed at the production-agent layer of its developer platform. The headline release is Claude Managed Agents, a hosted offering that bundles multiagent orchestration, scheduled “dreaming” runs, Outcomes telemetry, and webhooks into a managed surface that sits on top of the Claude API.

Alongside Managed Agents, Anthropic raised the rate limits on Claude Code and the Opus tier, signaling that the bottleneck for production agent workloads is no longer model availability so much as scaffolding. Self-hosted sandboxes — for teams that need to keep sensitive files, packages, and services in their own infrastructure — moved into public beta. MCP tunnels, which let an MCP server inside a private network expose tools to a Claude client without opening inbound ports, entered research preview.

The conference also brought a wave of legal-vertical infrastructure: more than 20 new MCP connectors for legal-research systems and 12 practice-area plugins covering research, contracts, discovery, matter management, and legal aid. The split between horizontal agent infrastructure and vertical packages is a deliberate one — Anthropic has positioned itself as both the model and the runtime, and these connectors are the runtime side of that pitch.

For operators currently building on the Claude API, the practical change is that the line between “use the API” and “deploy a multi-agent system” has narrowed. Where last year a production agent meant a half-dozen self-managed services around the model call, the May 6 announcements push more of that scaffolding into Anthropic’s plane.

The conference continues in London on May 20–21 and Tokyo on June 5–6.

Sources

• Anthropic Newsroom
• Live blog: Code w/ Claude 2026 — Simon Willison