Online checker that probes OpenAI-compatible API gateways for model mismatch, degraded capability and routing changes, with itemized evidence.
Free
Claude Managed Agents
Anthropic's hosted agent harness for long-running, asynchronous tasks: define an agent, pick a sandbox, start a session and stream results.
What is Claude Managed Agents?
Claude Managed Agents is a service on the Claude Platform that runs agents for you on infrastructure Anthropic manages. Anthropic describes two ways to build with Claude: the Messages API for custom agent loops and fine-grained control, and Managed Agents, a pre-built and configurable harness suited to long-running and asynchronous work. The quickstart walks through four steps. You create an agent (a model, system prompt, tools, MCP servers and skills), set up an environment (an Anthropic-managed cloud sandbox or a self-hosted one), start a session, and stream the agent's responses. Sessions can carry an optional hard spend budget, set when you create them, that pauses the work once list cost reaches your cap. The platform prices model tokens at each model's list rate, web searches at $10 per 1,000 searches, and session running time at $0.08 per hour. Requests use a beta header, so features and availability can change, and you need an Anthropic API key to call it.
Use cases
- Run long agent tasks in a cloud sandbox
- Hand off asynchronous work to Claude
- Build agents with MCP servers and skills
- Cap agent spend with session budgets
- Stream agent progress through an event stream
- Run agents in a self-hosted sandbox
Who it's for
Why use Claude Managed Agents
- You skip building the agent loop and the sandbox, which saves engineering time on long-running work.
- Sessions are asynchronous, so a task can keep going while your application does other things.
- A session budget gives you a ceiling on list cost before you hand an agent an open-ended job.
- Choosing between a managed cloud sandbox and a self-hosted one lets you fit your security setup.
- It sits next to the Messages API, so you can use direct prompting for some work and Managed Agents for the rest.
Claude Managed Agents features
- Pre-built agent harness running on Anthropic-managed infrastructure
- Agent definition with model, system prompt, tools, MCP servers and skills
- Anthropic-managed cloud sandbox or self-hosted environment
- Sessions with streamed agent responses
- Optional per-session spend budget with a USD hard cap
- Budgets that deployments apply to each session they start
- Idle state at budget limit that resumes when the cap changes
- SDK and ant CLI examples in several languages
- Setup help through Claude Code with /claude-api managed-agents-onboard
Claude Managed Agents integrations
How to get started with Claude Managed Agents
- Create an agent by defining the model, system prompt, tools, MCP servers and skills it should use.
- Create an environment that says where sessions run, either an Anthropic-managed cloud sandbox or a self-hosted sandbox.
- Start a session with your agent and environment, optionally passing a budget such as a maximum list cost in USD.
- Send the task and stream the agent's responses and events as it works.
- If the session reaches its budget it goes idle; raise or remove the cap to let the work continue.
Claude Managed Agents limitations
- Requests use a managed-agents beta header, so details can change
- A budget can only be attached when a session is created
- Budget enforcement happens between model requests, so spend can land slightly past the cap
- USD is the only currency supported for budgets
Claude Managed Agents pricing
Claude Managed Agents has no free plan. Plans start at $0.08 usage-based, according to the official pricing page on .
Claude Managed Agents plans
Session running time
$0.08 per hour
- Charged while a session runs
- Counts toward the session budget
Web search
$10 per 1,000 searches
- Counts toward the session budget
Claude Haiku 5.5
$0.10 /MTok input
Output $0.50 / MTok, prompts up to 100,000 tokens
- Model tokens at list price
Claude Sonnet 5.5
$2 /MTok input
Output $10 / MTok
- Model tokens at list price
Claude Opus 5.5
$4 /MTok input
Output $20 / MTok
- Model tokens at list price
Claude Fable 5.1
$10 /MTok input
Output $50 / MTok
- Model tokens at list price
Plans as listed on the official pricing page on . Prices can change and may vary by country or exclude tax. Check the official pricing page.
Claude Managed Agents alternatives
The main Claude Managed Agents alternatives listed on Debuly are OpenAI Agents SDK, LangGraph, CrewAI, Trigger.dev, Browser Use and Letta. The table compares pricing and platforms; order is A–Z.
| Tool | Pricing | Starting price | Platforms |
|---|---|---|---|
| Claude Managed Agents | Paid | $0.08, usage-based | API |
| Browser Use | Paid | $0.02, usage-based | API, Web, CLI |
| CrewAI | Open source | — | Web, API, CLI, Self-hosted |
| LangGraph | Open source | — | Self-hosted, Web, API |
| Letta | Freemium | $20/mo | CLI, Web, API |
| OpenAI Agents SDK | Open source | — | Python, macOS, Linux, Windows |
| Trigger.dev | Freemium | $10/mo | Web, API, CLI, Self-hosted |
Browser Use
Cloud browsers and hosted web agents for AI, plus an MIT-licensed open-source browser agent, billed pay-as-you-go from $0.02 per hour.
CrewAI
Multi-agent framework and platform for building AI agent crews in Python or a visual editor, then deploying and monitoring them.
Open sourceVisit CrewAI (official site)
LangGraph
Low-level open-source orchestration framework and runtime for building stateful, long-running AI agents as graphs.
Open sourceVisit LangGraph (official site)
Letta
Platform, formerly MemGPT, for building stateful AI agents with persistent memory, from a terminal app, SDK or API.
FreemiumVisit Letta (official site)
OpenAI Agents SDK
Open-source Python framework for building multi-agent workflows with tools, handoffs, guardrails, tracing, sandboxes and voice agents.
Open sourceVisit OpenAI Agents SDK (official site)
Trigger.dev
Open-source platform for durable AI agents and background workflows in TypeScript, with no timeouts, retries, queues and realtime.
Claude Managed Agents FAQ
How much does Claude Managed Agents cost?
The platform prices session running time at $0.08 per hour and web searches at $10 per 1,000 searches, plus model tokens at each model's list price. For example, Claude Sonnet 5.5 is $2 / MTok for input and $10 / MTok for output.
Is there a free plan?
Anthropic's docs do not describe a free plan for Managed Agents. Usage is priced by model tokens, web searches ($10 per 1,000) and session running time ($0.08 per hour).
How is it different from the Messages API?
The Messages API gives direct model access so you write your own agent loop. Managed Agents is a ready-made harness that runs in managed infrastructure for long-running, asynchronous work.
Can I limit how much a session spends?
Yes. You can set an optional budget with a maximum list cost in USD when you create the session. The session pauses and goes idle once the cap is reached, and raising or removing the cap resumes it.
Does Claude Managed Agents support MCP and skills?
Yes. An agent is defined by its model, system prompt, tools, MCP servers and skills, and sessions run in a cloud or self-hosted environment.
What are alternatives to Claude Managed Agents?
Alternatives include the OpenAI Agents SDK, LangGraph, CrewAI, Amazon Bedrock AgentCore and Google's Vertex AI Agent Builder.
Sources & verification
Facts on this page come from the official sources below, each with the date we read it. Written by Debuly; the vendor did not review this page.
Image credits
- Logo: from platform.claude.com · fetched · Vendor trademark, used only to identify the product; removed on request
- Cover: none collected yet, so a generated tile is shown.
Spot an error? Report an issue with this listing. We also remove a logo or image on the owner's request.
More in Developer platforms 4
All developer platforms toolsHardware-aware AI model optimization platform, NetsPresso, plus edge AI vision products for traffic, in-cabin monitoring and video analysis.
Unified OpenAI-compatible API for 500+ AI models with provider fallbacks, pay-as-you-go credits and a free tier
FreemiumSpatial generation company whose SuperSim rebuilds real-world sensor data into 3D simulations for training robots.
