18 backend blocks. 203 skills. One API. Let your AI agents build with production-ready infrastructure.
How it works
Same API your frontend uses. Now your agents can use it too.
Agent reads llms.txt or installs an MCP plugin to learn available blocks and skills.
Agent registers via AID protocol or uses your app-id and JWT. Two paths — admin-initiated or agent-initiated.
Agent calls skills — create users, upload files, manage data. 203 capabilities across 18 blocks.
Multiple ways for agents to integrate with 23blocks
Install 23blocks as Claude Code MCP servers. Each block is a standalone plugin with typed tools.
Standard HTTP endpoints for any agent or framework. JSON in, JSON out. Works with every LLM and agent runtime.
Machine-readable product summary following the llms.txt standard. Agents can discover capabilities automatically.
Two paths for agents to authenticate with your backend. Based on RFC 8628 device authorization.
Direct registration
Admin registers the agent with name, scopes, and Ed25519 public key
Agent receives JWT credentials and starts calling APIs immediately
RFC 8628 device flow
Agent requests registration and receives a user code (XXXX-XXXX)
Agent displays the code — admin approves it in the dashboard
Agent polls for approval and receives JWT tokens automatically
Install the Auth block as a Claude Code MCP plugin
$ claude mcp add-json "23blocks-auth" '{"type":"stdio","command":"npx","args":["23blocks-auth-mcp"]}'Everything about agents and 23blocks
AI Maestro coordinates multiple AI agents with persistent memory, agent-to-agent messaging, and a peer mesh network. Open source under MIT license.
Your backend is agent-ready
Sign up, get your API keys, and let your agents start building.
Get product updates, engineering posts, and new block announcements delivered to your inbox.