Deepmail

Your agent can search your email. It can't understand it.

Gmail MCP servers let agents read individual messages. But understanding email — knowing who matters, what relationships exist, what's happening in your life and business — requires investigation, cross-referencing, and memory. Agents can't do that in a single search.

Deepmail gives your AI agent 10 tools to investigate, analyze, and remember your email. It builds persistent knowledge that survives across sessions — so your agent actually knows you.

What your agent can do with Deepmail

You: "Who are my investors?"

Agent searches your Gmail for term sheets, fund communications, board emails.
Agent reads specific emails to understand deal details.
Agent cross-references names across rounds.
Agent builds a complete investor table — firms, key contacts, amounts, dates.

Next session, your agent already knows. No re-searching.

Quick start

pip install deep-email
deep-email auth

Opens your browser for Google consent. Read-only access. Done.

You may see "This app isn't verified" — click Advanced → Go to Deepmail (unsafe). This is standard for open-source tools. Deepmail only requests read-only email access.

Configure your agent

Add to your project's .mcp.json:

{
  "mcpServers": {
    "deepmail": {
      "type": "stdio",
      "command": "uvx",
      "args": ["deep-email"]
    }
  }
}

Or use the CLI:

deep-email init   # writes .mcp.json for you

Install the skill (teaches your agent how to use the tools)

npx skills add dennisonbertram/deepmail

Works with Claude Code, Cursor, Codex, GitHub Copilot, Windsurf, Gemini CLI, and 40+ other agents.

export GOOGLE_CLIENT_ID="your-id.apps.googleusercontent.com"
export GOOGLE_CLIENT_SECRET="your-secret"

Tools

How it works

1. Your agent reads the skill and learns how to investigate email topics
2. It searches your Gmail using search_emails — following leads, reading full emails when needed
3. It builds profiles using build_profile — extracting people, relationships, and context
4. Knowledge persists — who_is and about_me return cached results instantly in future sessions
5. It gets smarter — each investigation adds to the persistent knowledge base

All data stays on your machine. The MCP server runs locally. No cloud storage, no telemetry.

Examples

"Figure out my investors" — Agent searches for term sheets, fund emails, board communications. Builds a table of investors across rounds with key contacts and amounts.

"Who's on my team?" — Agent searches internal comms, standup emails, shared docs. Maps your org with roles and last-contact dates.

"Tell me about my relationship with [person]" — Agent searches all correspondence, synthesizes a relationship summary with timeline.

"What should I know before my meeting with [person]?" — Agent pulls recent threads, open items, relationship context.

License

AGPL-3.0 — free to use, modify, and distribute. If you run it as a hosted service, you must release your source under the same license. See LICENSE.

Security

• Read-only: Only gmail.readonly scope. Never sends, modifies, or deletes.
• Local-first: All data on your machine. No cloud, no telemetry.
• OAuth scoped: Minimum required permissions.
• Wipe anytime: reset_profile(confirm="yes") deletes all cached data.