Start with your agent
Connect Claude Code or another MCP client. Discover integrations, validate the graph and publish the workflow through the same API the web app uses.
Connect an agentTurn a process into a durable workflow. Connect your tools, bring your models, and let your agent build automations you can inspect and control.
runfabric auth loginCLI · MCP · SDKOpen-source engine. Provider-independent execution. Dev access by invitation.
v3 · publishedYour coding agent uses the schema, skills and tools to assemble a workflow. Runfabric gives every run a version, a history, and an explicit next step.
Connect Claude Code or another MCP client. Discover integrations, validate the graph and publish the workflow through the same API the web app uses.
Connect an agentBranches, loops, parallel work, timers and human approvals make the process explicit. Agentic steps decide how to use the tools you allow.
Explore the engineInspect the published graph, individual attempts, outputs and approvals. When an external result is uncertain, reconcile it with the evidence.
Explore the playgroundChoose OpenRouter or another compatible model endpoint. Connect HTTP APIs and MCP servers. Run private commands on a worker you control. Your model broker and your workflow engine have different jobs.
See how integrations workStart with the quickstart, connect a real integration, and follow your first run from trigger to result.
Start with the documentation