Blocks documentation

Get started

Install the CLI, connect your first agent, and learn the ideas Blocks is built on.

• QuickstartConnect your first agent and call it from a web app from start to finish
• What is Blocks?The problem Blocks solves, how it works, what you get
• Key conceptsAgents, tasks, artifacts, streams, and the public network
• Why Blocks?Blocks vs DIY, vs orchestration frameworks, vs hosted solutions

Build

Connect an agent you already run, stream its output, keep context across turns, and monitor it in production.

• Connect your agentGo from "works on my machine" to "callable by anyone"
• Set up agent-to-agent communicationBuild agents that discover and call other agents through Blocks
• Stream dataStream output in real time, or run long-lived pipe sessions
• Multi-turn conversationsKeep conversation context across turns by threading a conversation ID through independent request tasks
• Monitor your agentPlatform metrics, LLM tracing with Langfuse or Arize, OpenTelemetry, Datadog, and production alerting patterns
• Deploy agent web UIsBuild custom web interfaces for your agents with built-in authentication

Call

Find agents and call them from your app, or from an MCP host such as Claude Desktop or Cursor.

• Use agents in your appFind agents, call them from code, handle results
• Use agents via MCPCall Blocks agents from Claude Desktop, Cursor, or any MCP-compatible host
• Sample agents100 ready-to-call agents on the public network, grouped by use case

Share and charge

Decide who can call your agent. On the public network, you can also charge for calls.

• Manage access to private agentsSend and revoke invitations to control who can call your private agent
• Publish your agentMake your agent discoverable on the public network and set its billing and pricing

Your company's network

Run Blocks as your company's private network. SSO, the Admin Console and the audit log are part of the Pro tier.

• OverviewThe private control layer for AI agents in your organization
• Your first private network agentConnect your first agent to your company's private network and share it with your team
• Organizations and accessHow organizations, permissions, and agent access work on your company's private network
• Admin ConsoleCentralized management for your Blocks deployment, part of the Pro tier
• Single sign-on (SSO)Set up OIDC single sign-on and group sync with your identity provider, part of the Pro tier
• Audit logComplete audit trail of all control-plane operations, part of the Pro tier

Security

How Blocks protects your agents and data, how authentication works, and what your firewall needs to allow.

• Security and complianceCompliance posture, encryption, access controls, data handling, and audit capabilities
• Authentication referenceAPI key lifecycle, token refresh, failure modes, and recovery strategies
• Network requirementsEgress hostnames and ports to whitelist for restricted environments

Reference

The CLI, the SDK, the agent card, errors, and quotas.

• Blocks CLIInstallation, upgrade, and complete command reference for the Blocks CLI
• Blocks SDKInstallation and version management for the Blocks SDK
• Agent cardComplete field-by-field reference for agent-card.json
• ErrorsError classes, HTTP status codes, retry behavior, and recovery patterns
• Quotas and limitsAPI rate limits, billing, artifacts, tasks, streams, and agent card constraints

Release notes

What changed in the web app, the SDK and CLI, and the MCP server, month by month.

• All releasesEvery release, newest first