Anthropic ships five developer updates to Claude Managed Agents: streaming deltas, session overrides, and richer webhooks
Anthropic released five platform updates to Claude Managed Agents on June 30, 2026, covering real-time event streaming, per-session configuration flexibility, and a more complete webhook system that eliminates polling for agent lifecycle events.
What's new
Event deltas for session streams. Managed Agents session event streams now support opt-in event deltas via the event_deltas[] query parameter on GET /v1/sessions/{session_id}/events/stream. Two new event types — event_start and event_delta — stream a preview of agent message text as it is generated, before the complete agent.message event arrives. This gives developers low-latency access to partial output without waiting for a full message.
Backward pagination for session listing. GET /v1/sessions now returns a prev_page cursor alongside the existing next_page cursor. Passing it as the page parameter navigates to the previous page, enabling bidirectional traversal of session history.
Per-session agent config overrides. When creating a session, developers can now supply an agent field with type: "agent_with_overrides" to replace the underlying agent's model, system prompt, tools, MCP servers, or skills for that session only. The agent definition itself remains unchanged. This makes it straightforward to run the same agent logic with different models or restricted tool sets without forking the agent.
Vault credential injection control. Environment variable credentials in Managed Agents vaults now support an injection_location setting that controls whether the credential value is substituted into the agent's outbound request headers, the request body, or both at egress. Previously, injection behavior was fixed.
Expanded webhooks. Webhooks for Managed Agents now cover the full agent, deployment, and deployment run lifecycle — including newly published agent versions, paused deployments, and failed scheduled runs. Subscribers no longer need to poll the API to detect these state changes.
Context
Claude Managed Agents — Anthropic's hosted platform for running persistent Claude agents at scale — has been expanding steadily since its launch earlier this year. Prior updates added scheduled deployments, multi-agent orchestration, and self-hosted sandbox support. These June 30 changes are squarely aimed at production deployments: streaming deltas reduce latency for interactive interfaces, per-session overrides enable multi-tenant or A/B testing scenarios, and the richer webhook coverage makes it practical to build event-driven operations pipelines on top of agents without resorting to polling.
The same June 30 release also introduced Claude Sonnet 5 (claude-sonnet-5), the next generation of the Sonnet family, at introductory pricing of $2/$10 per million tokens through August 31, 2026.
Why it matters
For developers running agents in production, these five updates address recurring operational friction points. The ability to override a session's model or tools without touching the agent definition is particularly useful for organizations that maintain a canonical agent but need to route different user segments to different compute tiers or capability sets. The webhook expansion moves Managed Agents closer to a complete event-driven platform: instead of polling the API to learn that a scheduled run failed or a new agent version was published, subscribers can react immediately.
Collectively, these changes signal that Anthropic is treating Managed Agents as a first-class production surface rather than a preview feature — adding the operational tooling that enterprise deployments require.
Corroborating sources
- Platform.claude
https://platform.claude.com/docs/en/release-notes/api
“You can react to a newly published agent version, a paused deployment, or a failed scheduled run without polling.”