Skip to main content

Agents on Every MCP Server You Enable

· 4 min read
Tony Lewis
MCPBundles
Ask AI:
PerplexityClaudeChatGPTGoogleGrok

TL;DR​

  • You can attach an A2A agent to any MCP server in your workspace—the same catalog surface you use from Claude or Cursor—not only from the Hub.
  • Agent instructions are enough to run manually or from chat; a heartbeat checklist is optional and mainly for recurring scheduled runs.
  • On a bundle MCP connection, agent tools use the same names as the Hub (get_agents, upsert_agent, trigger_agent_run, and the rest).

It's Tuesday standup and someone asks whether yesterday's Google Ads spend spiked for a client you do not have open in another tab. You already have the MCP server in Cursor. What you want is a named assistant on that server that you can wake on demand or on a clock.

That is the workflow we optimized this week.

Same server, not a second stack​

For a while the story was: agents on the Hub, product tools on each server. That kept webhook and schedule work in one place while we chased edge cases in async task chains.

It also meant "my Shopify monitor" lived somewhere else mentally. You would enable Shopify in chat, then hop to the Hub to define who runs overnight inventory checks.

Agent tools now ship on every bundle MCP endpoint. Point Cursor at google-ads and the tool list includes campaign reads plus upsert_agent. The agent is bound to that server automatically. No extra wiring step.

Hub and bundle share one implementation and the same wire names, so CLI scripts do not fork.

Instructions first, checklist when you want rhythm​

Readiness no longer forces a second markdown file before your first run.

Agent instructions cover role, tone, and how to use tools on that server. That is enough for Run now from the dashboard, Hub, bundle MCP, or an external A2A client.

Add a heartbeat checklist when you want scheduled passes with the same steps each time: scan inbox, pull yesterday's spend, flag disapproved ads. A one-off "do what my instructions say" agent can skip it.

We tried the lean path on Google Ads: one agents.md, no checklist, triggered from bundle MCP. One tool call to read spend, a short reply, done.

From chat, without a new UI​

An ops lead already has Google Ads in Claude Code. They declare the assistant once: Create an agent on this server called Morning PPC. Read yesterday's spend, flag disapproved ads, never enable campaigns without asking.

When spend jumps before lunch, the same thread gets: Run Morning PPC now and paste the summary here.

Turn on Activate and an interval when you want hourly standup prep or a nightly spend check. Webhooks and public agent pages behave as before; see Creating A2A Agents.

Skills stay domain-only​

Calling get_skill on Google Ads or Shopify still returns product guidance: entities, workflows, gotchas. It does not teach agent creation. That stays on the Hub skill and in docs so hundreds of server skills do not repeat platform setup.

Where to go​

MCP gave you the toolbox on each server. Agents are the teammate who already knows which drawer to open, on that server, on a clock or when you ping them.