undefined

Build email automation with Claude and Python. Covers drafting replies, classifying inbound email, extracting action items, bulk processing with Batch API, and Gmail/SMTP integration.

Claude Email Automation Python

Automate email workflows with Claude — draft replies, classify inbound messages, extract action items, and bulk-process with the Anthropic Batch API at 50% cost.

Draft an email reply

import anthropic

client = anthropic.Anthropic()

def draft_reply(original_email: str, context: str = "") -> str:
    """Draft a professional reply to an incoming email."""
    system = (
        "You are an expert email assistant. Draft concise, professional replies. "
        "Match the tone of the original. Output only the email body — no subject line, "
        "no greeting/sign-off unless explicitly asked."
    )
    prompt = f"Original email:\n\n{original_email}"
    if context:
        prompt += f"\n\nAdditional context: {context}"

    response = client.messages.create(
        model="claude-haiku-4-5-20251001",
        max_tokens=512,
        system=system,
        messages=[{"role": "user", "content": prompt}],
    )
    return response.content[0].text

# Usage
reply = draft_reply(
    original_email="Hi, can we move our 3pm meeting to 4pm tomorrow?",
    context="I'm free at 4pm. Keep it brief.",
)
print(reply)

Classify inbound email (support routing)

import json

CATEGORIES = ["billing", "technical_support", "sales_inquiry", "general", "spam"]

def classify_email(subject: str, body: str) -> dict:
    """Classify email and return category + priority + summary."""
    response = client.messages.create(
        model="claude-haiku-4-5-20251001",
        max_tokens=256,
        system="You are an email classifier for a SaaS company. Return valid JSON only.",
        messages=[{
            "role": "user",
            "content": (
                f"Subject: {subject}\nBody: {body[:2000]}\n\n"
                f"Classify into one of {CATEGORIES}. "
                "Return JSON: {{\"category\": str, \"priority\": \"high|medium|low\", "
                "\"summary\": str (max 20 words), \"auto_reply\": bool}}"
            ),
        }],
    )
    return json.loads(response.content[0].text)

result = classify_email(
    subject="URGENT: Can't log in — production down",
    body="Our entire team is locked out. This is blocking us.",
)
# {"category": "technical_support", "priority": "high", "summary": "Team locked out, production down", "auto_reply": false}

Extract action items from email threads

def extract_action_items(email_thread: str) -> dict:
    """Extract tasks, deadlines, and owners from an email thread."""
    response = client.messages.create(
        model="claude-sonnet-4-6",
        max_tokens=512,
        system="Extract action items from email threads. Return valid JSON only.",
        messages=[{
            "role": "user",
            "content": (
                f"Email thread:\n\n{email_thread}\n\n"
                "Return JSON: {{\"action_items\": [{{\"task\": str, \"owner\": str|null, "
                "\"deadline\": str|null, \"priority\": \"high|medium|low\"}}]}}"
            ),
        }],
    )
    return json.loads(response.content[0].text)

thread = """
From: Alice: Can you review the contract by Friday?
From: Bob: Sure, I'll handle it. Can you send the vendor details too?
From: Alice: Yes, I'll send them by EOD Wednesday.
"""
items = extract_action_items(thread)
# action_items: [{task: "Review contract", owner: "Bob", deadline: "Friday", priority: "high"}, ...]

Bulk email processing with Batch API (50% cost)

def bulk_classify_emails(emails: list[dict]) -> list[dict]:
    """Classify a list of emails in one Batch API call at 50% cost."""
    requests = [
        {
            "custom_id": f"email-{i}",
            "params": {
                "model": "claude-haiku-4-5-20251001",
                "max_tokens": 128,
                "system": "Classify this email. Return JSON: {category, priority, auto_reply}.",
                "messages": [{
                    "role": "user",
                    "content": f"Subject: {e['subject']}\n{e['body'][:500]}",
                }],
            },
        }
        for i, e in enumerate(emails)
    ]

    batch = client.messages.batches.create(message_batch=requests)
    print(f"Batch {batch.id} submitted — poll or use webhook for results")
    return batch  # poll batch.id for results when processing_status == "ended"

Gmail integration via google-api-python-client

pip install google-api-python-client google-auth-httplib2 google-auth-oauthlib anthropic

from googleapiclient.discovery import build
from google.oauth2.credentials import Credentials
import base64, email as email_lib

def fetch_and_reply(creds_path: str, max_emails: int = 5):
    """Fetch unread Gmail messages and draft Claude replies."""
    creds = Credentials.from_authorized_user_file(creds_path)
    service = build("gmail", "v1", credentials=creds)

    results = service.users().messages().list(
        userId="me", q="is:unread", maxResults=max_emails
    ).execute()

    for msg_meta in results.get("messages", []):
        msg = service.users().messages().get(
            userId="me", id=msg_meta["id"], format="full"
        ).execute()
        # Decode body
        data = msg["payload"]["body"].get("data", "")
        body = base64.urlsafe_b64decode(data + "==").decode("utf-8", errors="ignore")

        subject = next(
            (h["value"] for h in msg["payload"]["headers"] if h["name"] == "Subject"), ""
        )
        reply = draft_reply(original_email=f"Subject: {subject}\n\n{body[:1000]}")
        print(f"Draft reply for '{subject}':\n{reply}\n")

Claude email automation vs alternatives

Key patterns: (1) Use claude-haiku-4-5-20251001 for classification and short drafts — it is 20× cheaper than Sonnet and fast enough for real-time routing. (2) Use the Batch API for bulk processing (nightly digest, backlog clearance) — 50% cost with async delivery. (3) Keep the original email body under 2000 chars for classification prompts; use Sonnet for long-thread extraction. For cost estimates, use the Claude API Cost Calculator. For structured data extraction from email attachments (PDFs, invoices), see the data extraction guide.

Free tools

More examples