Add Claude Managed Agents integration docs#402
Conversation
Document running Anthropic's hosted Managed Agents harness against Kernel cloud browsers: durable resources vs ephemeral sessions, the coordinator/ worker topology, the environment-vs-credential networking model, and a step-by-step quickstart. Add it to the integrations nav and overview. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
|
Preview deployment for your docs. Learn more about Mintlify Previews.
💡 Tip: Enable Workflows to automatically generate PRs for you. |
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
Add a "what this enables" thesis paragraph framing the runtime/browser split, a benefits section near the top, and note the auto-set beta header. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
There was a problem hiding this comment.
Risk assessment: Low.
The current diff adds a new Claude Managed Agents documentation page and two navigation links (docs.json, integrations/overview.mdx). There are no application code, infrastructure, auth, billing, permissions, or data-path changes. I’m rating it Low rather than Very Low because the new page includes copyable agent prompts, CLI commands, beta API details, and credential-handling guidance, so documentation mistakes could affect users who follow it.
The PR was already approved before this update, and the update since that approval only adds explanatory docs copy and beta-access clarification within the same page, so I don’t see a risk increase that would require dismissing the prior approval. I found no CODEOWNERS file indicating Codeowners review is required.
Sent by Cursor Automation: Assign PR reviewers
| The example code below is TypeScript. The same API is exposed through the Anthropic SDKs for other languages under their `beta` namespace as Managed Agents rolls out. | ||
| </Note> | ||
|
|
||
| ## Quickstart |
There was a problem hiding this comment.
@rgarcia should we also link to the cookbook? would require making the cookbook public
There was a problem hiding this comment.
will deploy as-is to get back to hema before week start, can add later once cookbook is public
There was a problem hiding this comment.
I'll clean up the cookbook and open source it tomorrow.
The quickstart uses only the Anthropic SDK and process.env — none of the recent Node APIs that would require a higher floor. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
There was a problem hiding this comment.
Risk assessment: Low.
The current diff adds a new Claude Managed Agents documentation page and links it from docs.json and integrations/overview.mdx. There are no application code, infrastructure, auth, billing, permissions, schema, or production data changes. I’m rating it Low rather than Very Low because the new docs include copyable agent prompts, CLI commands, beta API details, and credential-handling guidance, so inaccurate guidance could affect users following the page.
The PR is already approved, so I’m not adding another approval. I re-evaluated the synchronized update and found it only changes the Node.js prerequisite from 22+ to 18+ within the same docs page, which doesn’t increase the risk level. I found no CODEOWNERS file indicating Codeowners review is required.
Sent by Cursor Automation: Assign PR reviewers
juecd
left a comment
juecd
left a comment
There was a problem hiding this comment.
Nothing blocking in my end for launch 🔥 I might look at it later with fine tooth comb
| - **No infrastructure to run**: Anthropic hosts the agent loop and Kernel hosts the browsers — no cold starts, container orchestration, or Chromium version pinning on your side. | ||
| - **Parallel by default**: fan out one browser per subagent and run them concurrently, so a single coordinator can cover many pages at once. | ||
| - **Stealth, non-headless browsing**: [stealth mode](/browsers/bot-detection/stealth) drives real Chromium that renders and behaves like a human visitor, not a flagged bot. | ||
| - **Persistent session state**: carry cookies and logins across runs with [Profiles](/auth/profiles), so agents resume where they left off. |
There was a problem hiding this comment.
I might call out Managed Auth specifically - Claude Managed Agents can seamlessly access browsers on behalf of users using our secure, permissioned abstraction
| - **Stealth, non-headless browsing**: [stealth mode](/browsers/bot-detection/stealth) drives real Chromium that renders and behaves like a human visitor, not a flagged bot. | ||
| - **Persistent session state**: carry cookies and logins across runs with [Profiles](/auth/profiles), so agents resume where they left off. | ||
| - **Built-in observability**: watch agents drive their browsers live with [Live View](/browsers/live-view), or review a run afterward with [session replays](/browsers/replays). | ||
| - **Clean separation of state**: Managed Agents holds the conversation and tool outputs; Kernel holds the page, cookies, and downloads — so you can inspect each side on its own. |
There was a problem hiding this comment.
Yes, I like this esp. given it's a big focus for CMA's architectural design.
| The example code below is TypeScript. The same API is exposed through the Anthropic SDKs for other languages under their `beta` namespace as Managed Agents rolls out. | ||
| </Note> | ||
|
|
||
| ## Quickstart |
| - Learn about [stealth mode](/browsers/bot-detection/stealth) for reliable, non-headless browsing | ||
| - Use [Playwright Execution](/browsers/playwright-execution) to run structured Playwright from the CLI | ||
| - Debug runs with [live view](/browsers/live-view) | ||
| - Persist browser state across sessions with [Profiles](/auth/profiles) |
There was a problem hiding this comment.
Link to Managed Auth instead of Profiles
|
Firetiger deploy monitoring skipped This PR didn't match the auto-monitor filter configured on your GitHub connection:
Reason: This PR adds documentation to the To monitor this PR anyway, reply with |
There was a problem hiding this comment.
Risk assessment: Low.
The current diff adds a new integrations/claude-managed-agents.mdx documentation page and links it from docs.json and integrations/overview.mdx. I don’t see application code, infrastructure, auth, billing, permissions, schema, or production data-path changes.
I’m rating it Low rather than Very Low because the new guide includes copyable agent prompts, CLI commands, beta API details, and credential-handling guidance; mistakes there could affect users who follow the documentation, but the blast radius remains limited to docs. The PR is already approved, so I’m not adding another approval. I found no CODEOWNERS file indicating Codeowners review is required.
Sent by Cursor Automation: Assign PR reviewers
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 1 potential issue.
❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.
Reviewed by Cursor Bugbot for commit 7b3b9ac. Configure here.
| - **Parallel by default**: fan out one browser per subagent and run them concurrently, so a single coordinator can cover many pages at once. | ||
| - **Stealth, non-headless browsing**: [stealth mode](/browsers/bot-detection/stealth) drives real Chromium that renders and behaves like a human visitor, not a flagged bot. | ||
| - **Managed Authentication**: Claude Managed Agents can seamlessly access browsers on behalf of users using Kernel's secure, permissioned abstraction of user identity | ||
| - **Persistent session state**: carry cookies and logins across runs with [Profiles](/auth/profiles), so agents resume where they left off. |
There was a problem hiding this comment.
Persistent session wording
Medium Severity
The benefits bullet uses the heading “Persistent session state” and points to Profiles, which conflicts with deprecating persistent browser sessions and with the same page’s Managed Auth positioning elsewhere.
Triggered by learned rule: Flag references to persistent sessions (deprecated feature)
Reviewed by Cursor Bugbot for commit 7b3b9ac. Configure here.
There was a problem hiding this comment.
Risk assessment: Low.
The current diff adds a new integrations/claude-managed-agents.mdx documentation page and links it from docs.json and integrations/overview.mdx. There are no application code, infrastructure, auth logic, billing, permissions, schema, or production data-path changes.
I’m keeping this at Low rather than Very Low because the page includes copyable agent prompts, CLI commands, beta API details, and credential-handling guidance; inaccuracies could affect users who follow the docs, but the blast radius is limited to documentation. Since the prior assessment, the synchronized update only adds a Managed Authentication benefit bullet and changes a next-step link from Profiles to Managed Auth, so I don’t see a risk increase that would require dismissing approval.
The PR is already approved, so I’m not adding another approval. I found no CODEOWNERS file indicating Codeowners review is required.
Sent by Cursor Automation: Assign PR reviewers
3 participants
Summary
Adds an
integrations/claude-managed-agentspage documenting how to run Anthropic's hosted Claude Managed Agents harness against Kernel cloud browsers — the analog to the existing Claude Agent SDK page, but for the hosted agent loop (agents, sessions, environments, vaults on Anthropic's side, reaching Kernel over the network).The page covers:
KERNEL_API_KEY, never plaintext in the container or model context).Also wires it into the Integrations nav (
docs.json) right after Claude Agent SDK, and into the integrations overview list.Notes for reviewers
agent_toolset_20260401and themanaged-agents-…-research-previewheader are shown as-is; revisit at GA.<Note>points to other-language SDKs as Managed Agents rolls out. Happy to add a Python tab once that surface is stable.docs.jsonvalidated as JSON.Test plan
mintlify devand confirm the page renders + appears in the Integrations nav🤖 Generated with Claude Code
Note
Low Risk
Documentation-only changes with no runtime, auth, or data-path code; main risk is publishing beta/preview API details before GA.
Overview
Adds
integrations/claude-managed-agentsdocumentation for pairing Anthropic’s hosted Claude Managed Agents harness with Kernel cloud browsers (contrast with the self-hosted Claude Agent SDK page).The new guide explains durable vs ephemeral resources (environment, vault, agents vs sessions), driving browsers via
kernelCLI inside the sandbox, vaultKERNEL_API_KEYhandling with host allowlists, and a TypeScript quickstart (environment → vault credential → coordinator/worker agents withweb_fetchdisabled → session streaming → teardown). It also notes beta/preview API identifiers and early-access requirements.Navigation: registers the page in
docs.json(Integrations, after Claude Agent SDK) and links it fromintegrations/overview.mdx.Reviewed by Cursor Bugbot for commit 7b3b9ac. Bugbot is set up for automated code reviews on this repo. Configure here.