Running Two Agents at Once: Claude for Communication, Cursor for Code

Claude manages your email and communication. Cursor manages your codebase and dev tools. Both run through Stow — same policy layer, separate Security Baselines, zero overlap.

Claude and Cursor are optimized for different things. Claude is conversational, context-aware, great at email and writing. Cursor is code-native, file-aware, great at multi-file edits and dev tool orchestration. Running both through Stow gives you a unified permission layer with separate Security Baselines — they share infrastructure but never share access.

Why Separate Agents Instead of One

One agent for everything

Single Security Baseline (breaks when you switch contexts)
Claude sees dev tools it doesn't need
Cursor has email access it shouldn't have
One credential to lose

Two agents, scoped access

Independent Security Baselines per network context
Claude only sees communication services
Cursor only sees dev tools
Credentials are isolated — one breach doesn't expose both

The Recommended Service Split

Claude Desktop

Auth: OAuth

Gmail — read, draft, send with approval
Slack — read, DM freely, channels with approval
Notion — read and create pages
Google Drive — read and create docs
Calendly — list events, share links

Cursor

Auth: SSE + credentials

GitHub — create branches and PRs
Vercel — view logs, trigger builds with approval
Neon — create branches, list databases
Slack — post to dev channels with approval

Notice Slack appears in both — but for different purposes. Claude uses it for communication. Cursor uses it for dev notifications. Each agent only sees the Slack operations relevant to its role, and each has independent approval settings.

Setting Up Claude Desktop in Stow

In Stow, go to AI Agents → + Add Agent, name it "Claude Desktop" and select the Claude icon
Select Claude Desktop as the implementation method — copy the base URL
In Claude Desktop, go to Customize → Connectors → Add Custom Connector
Paste https://mcp.stowprotect.com and click Add → Connect
OAuth flow completes automatically — Security Baseline is captured on first request

Setting Up Cursor in Stow

In Stow, go to AI Agents → + Add Agent, name it "Cursor" and click New Client Secret
Copy the Agent ID and Secret
In Cursor Settings (Cmd/Ctrl+Shift+J), go to Tools & MCP → + New MCP Server
Paste the SSE config with your credentials:

{
  "mcpServers": {
    "stow": {
      "type": "sse",
      "url": "https://mcp.stowprotect.com/sse?agent_id=YOUR_AGENT_ID&agent_secret=YOUR_SECRET"
    }
  }
}

Independent Security Baselines

This is the key advantage of two separate agents. Claude Desktop typically runs from your home or office network. Cursor might run from a different machine, a CI environment, or a remote dev container. They establish independent baselines on first connect:

Claude Desktop baseline

Captures the network context of your primary device — home office, laptop, etc.

Cursor baseline

Captures the network context of your dev machine — which may be different, especially with remote dev environments.

If Cursor's baseline deviates (remote VM, new network), only Cursor's requests are affected. Claude continues operating normally. You rotate Cursor's secret independently.

A Day in the Dual-Agent Workflow

Morning

Claude reads your overnight emails, drafts replies, and flags urgent items in a Notion summary

Mid-morning

Cursor works on the feature branch — creating files, running tests, committing changes

Noon

Cursor opens a PR; you get a Slack DM summary from Claude with what changed

Afternoon

Claude handles follow-up emails and meeting prep while Cursor addresses PR feedback

End of day

Cursor posts a completion summary to #engineering; Claude drafts a team update for tomorrow

Two Agents. One Policy Layer.

Claude handles communication. Cursor handles code. Neither touches what the other owns.

Stow Security Team

April 19, 2026