← agentid.live
Multi-agent coordination

Two agents.
One identity.
No orchestration layer.

AgentID's handoff() tool lets one agent pass work to another through persistent shared memory. No message queues. No API calls between agents. Just a structured note in memory that the next agent reads on startup.

🔍
@researcher_agent
Researcher
active
Session 1 — Gathers information, writes findings to memory, then hands off
Persona: dev-blog-identity
✍️
@writer_agent
Writer
waiting
Session 2 — Reads the handoff on startup and picks up exactly where research ended
Persona: dev-blog-identity
TRANSCRIPTResearch → Handoff → Write
Session 1 — @researcher_agent in Claude Desktop
Research the top patterns developers are using MCP servers for in 2025. Take notes in memory. When done, hand off to @writer_agent to write the blog post.
🔍
@researcher_agent
report_activity
type: "task.started" · "Researching MCP patterns for developer blog post"
start_mission
"Blog: MCP Patterns 2025" · goal: research + write developer blog post
↳ Mission created. 2 agents on this identity: @researcher_agent, @writer_agent
write_memory
key: "blog_outline" · 143 words — themes, structure, target audience
write_memory
key: "section_identity" · 97 words — persistent identity across sessions
write_memory
key: "section_multi_agent" · 112 words — coordination patterns, handoff mechanics
update_status
state: "done" · "Research complete — 3 sections outlined, all notes in memory"
🔍
@researcher_agent
handoff
→ @writer_agent · 3 next steps · summary: research complete, notes in memory

Done. Research is wrapped up — outline and section notes are in memory. @writer_agent will pick this up when they start their next session.

⇌Handoff written to shared memory
From → To
@researcher_agent → @writer_agent
Summary
Research complete. 3 sections outlined, notes saved to memory keys: blog_outline, section_identity, section_multi_agent
Next steps
  1. Read memory keys: blog_outline, section_identity, section_multi_agent
  2. Write 800-word post targeting developers building with Claude
  3. Use concrete examples — real tool calls, not abstractions
Session 2 — @writer_agent in Cursor (separate session)
New session. @writer_agent starts up. No user message needed — the agent reads the mission first.
✍️
@writer_agent
read_mission
checking for active missions and handoffs addressed to me…
📋Handoff received via read_mission()
Mission: Blog: MCP Patterns 2025 — 1 active agent, 1 completed
YOUR ACTION: This handoff is addressed to you (@writer_agent). Pick up from the next steps above. Call update_status() once you start.
✍️
@writer_agent
update_status
state: "working" · "Picked up handoff from @researcher_agent. Reading notes."
read_memory
key: "blog_outline"
↳ # MCP Patterns 2025 — Outline: 3 sections, 800 words, dev audience...
read_memory
key: "section_identity"
↳ Persistent identity: The biggest friction in agent dev is every session starts fresh...
read_memory
key: "section_multi_agent"
↳ Multi-agent coordination: The handoff primitive passes work with a structured summary...
report_activity
type: "task.completed" · "Blog post written — 863 words"

# MCP Patterns in 2025: What's Actually Working

The biggest shift in how developers are using MCP servers isn't about tools — it's about memory. Here's what's working in production...

[863 words — draft complete]

1
Same persona, different agents
Both agents connect with different handles but share one persona. Shared memory, shared mission state.
2
Handoff writes to memory
handoff() stores a structured note — summary + next_steps + recipient — at a reserved memory key.
3
Pickup is automatic
Any agent that calls read_mission() sees active handoffs addressed to them. No polling. No webhooks.
Set this up in 5 minutes

Add both agents to your claude_desktop_config.json. Each gets its own handle, both share one persona.

{
  "mcpServers": {
    "researcher": {
      "url": "https://agentid.live/api/mcp/researcher_agent",
      "headers": { "Authorization": "Bearer YOUR_API_KEY" }
    },
    "writer": {
      "url": "https://agentid.live/api/mcp/writer_agent",
      "headers": { "Authorization": "Bearer YOUR_API_KEY" }
    }
  }
}

Both agents will be connected to the same persona automatically. Get your API key at agentid.live/app/developers.

Build your own multi-agent pipeline →
Free to start · No credit card required · See full examples on GitHub