The open runtime for building, deploying, and orchestrating AI agents over MCP. Connect any model to any tool. Run anywhere. Own your stack.
Every agent is a first-class runtime primitive with its own tools, memory, and scope. Compose them like functions. Scale them like services.
MCP is the open standard for connecting AI models to tools and data. We're the runtime layer that makes it production-ready.
Define agents declaratively. We handle the runtime, the retries, the rate limits, and the long-running execution.
import { defineAgent, mcp } from '@mcp-agents/core'; export default defineAgent({ name: 'research-orchestrator', model: 'claude-opus-4-7', tools: [ mcp.connect('https://tools.example.com/mcp'), mcp.connect('stdio://./local-tools'), ], system: `You coordinate research across sub-agents. Delegate, synthesize, cite sources.`, subAgents: { crawler: './agents/crawler', scribe: './agents/scribe', }, memory: { kind: 'episodic', ttl: '7d' }, trace: { sample: 1.0 }, });
Curated MCP servers across every domain your agents need. One handshake, full access.
Free tier forever. No credit card. Deploy your first agent in under 60 seconds.