{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreibucojp4u3rpughju6wy3r4cxvnl4xjdbwfaviomct6yidkprtkbm",
    "uri": "at://did:plc:lk3jfj3zq4k4wxnk474axylu/app.bsky.feed.post/3mgs7f6rbigg2"
  },
  "path": "/t/agentrouter-run-specialized-agents-in-real-time/1376363#post_1",
  "publishedAt": "2026-03-11T14:58:00.000Z",
  "site": "https://community.openai.com",
  "tags": [
    "(click for more details)"
  ],
  "textContent": "Hey Agent-Builders\n\nA while ago I was building a **scientific research agent** in n8n and crewai.\n\nI kept asking myself:\n\n> How can I make my agent more capable without turning it into a giant, over-engineered workflow?\n\nThe idea of **multi-agent systems and external expert agents** kept coming up for me.\n\nSo I built something to experiment with that idea:\n\n## AgentRouter\n\nA small MCP-based tool that lets your agent **delegate tasks to specialized external agents in real time**.\n\nInstead of building one massive agent with dozens of tools, your agent can now **hire expert sub-agents when it needs them.**\n\nExamples of agents it can call:\n\n• browser navigation agents\n• academic research agents\n• code execution agents\n• other specialized AI agents\n\nYour agent simply sends a request — and gets the result back.\n\n* * *\n\nHow it works\n\nThe router connects to the `agent-router.org` platform via an **MCP Client Tool** and discovers, calls and recives results from specialised agents autonomsly.\n\nSetup is pretty simple:\n\n`{`\n`“mcpServers”: {`\n`“a2a-platform”: {`\n`“command”: “npx”,`\n`“args”: [`\n`“-y”,`\n`“mcp-remote”,`\n`“https://a2a-backend-196084590575.europe-west1.run.app/mcp/sse?apiKey=sk_a2a_…”`\n`]`\n`}`\n`}`\n`}`\n\nAutomation Platforms (n8n, make etc.) (click for more details)\n\nYou can get your **free API key** here:\n`agent-router.org/dashboard`\n\nAdditionally you should add this system prompt to your agent, so he uses the agent optimal:\n\nSystem Prompt (click for more details)\n\n* * *\n\n## Why I built this\n\nWhen building agents in n8n, the builder often has to **manually orchestrate everything**.\n\nMy goal with this tool is to make agents:\n\n• more capable\n• more modular\n• able to collaborate with other agents\n\n* * *\n\n## I’d love feedback from the community\n\nThis is still an early experiment and I’d really appreciate your thoughts.\n\nSome questions I’m currently exploring:\n\n  * Do **you build agents** that would profit from external agents and delegations?\n\n  * Is this tool **actually helping your Agent** or is it a nice to have thing?\n\n  * What types of **specialized agents** would be most useful for you?\n\n  * Would you **pay** for this?\n\n\n\n\n* * *\n\nIf people are interested I can share:\n\n• the architecture\n• integrations for custom agents\n• example workflows / use cases\n\nCurious what you think",
  "title": "AgentRouter — Run specialized agents in real time"
}