Claude Managed Agents

Claude Managed Agents is Anthropic's hosted agent runtime: you define model, system prompt, tools, and MCP servers once, then run sessions in managed sandboxes. Thirdfy MCP plugs in as a remote MCP server so your agent can discover the action catalog, preflight, and execute intents without shelling out to the CLI.

Anthropic docs

Start with Anthropic's guides, then return here for Thirdfy MCP wiring.

Console uses the same /v1/agents and /v1/sessions API as the SDK. You can copy the generated request from Console when you move to code.

Connect Thirdfy MCP

In Claude Console (or in your agent's mcp_servers array):

1. Create or open a Managed Agent in Claude Console.
2. Add an MCP server (streamable HTTP):
   • Name: thirdfy (or any unique name you reference in tools)
   • URL: https://mcp.thirdfy.com/mcp
3. Enable MCP tools on the agent (an mcp_toolset that references your server name). See Anthropic MCP connector.
4. Authenticate Thirdfy: store your agent API key in a Claude vault credential (static_bearer) and attach that vault when you start a session. Onboarding path: MCP onboarding or Connect.

Thirdfy endpoint details: MCP Connect.

First Thirdfy tool calls

After the connector is live:

1. Discovery: getActionsCatalog, getProviderActions
2. Onboarding: describeOnboardingPath, startEmailOnboarding, completeEmailOnboarding (issues agent API key and execution wallets)
3. Preflight: run before live writes
4. Execution: executeIntent, walletExecute, or agentRun

Full tool list: Thirdfy MCP.

CLI vs Claude Managed Agents

Both use the same Thirdfy policy layer and action catalog.

Related

• Agent CLI
• MCP
• Quick start
• Delegation and protection