Inbox Zero

Exposes email data through the Model Context Protocol (MCP) standard, allowing Claude and other LLM clients to query, search, and retrieve email messages using semantic search and structured filtering. Implements MCP resource handlers that translate email queries into database lookups, enabling LLMs to access email context without direct API integration or authentication management.

Exposes email operations (send, archive, delete, label, snooze) as MCP tool definitions that LLMs can invoke directly. The server implements tool handlers that validate action parameters, apply business logic (e.g., prevent accidental mass deletion), and execute changes against the email backend, enabling AI agents to take autonomous email management actions.

Abstracts differences between email providers (Gmail, Outlook, IMAP) behind a unified interface, translating provider-specific APIs and authentication mechanisms into consistent MCP resource and tool definitions. The server handles provider-specific label mappings, rate limiting, and protocol differences transparently, allowing LLM clients to interact with any supported email provider using identical MCP calls.

Parses raw email messages (MIME format, HTML, plain text) into structured data, extracting sender, recipient, subject, body, attachments, and metadata. Implements HTML-to-text conversion, MIME decoding, and optional NLP-based entity extraction (dates, action items, decision points) to make email content machine-readable for LLM analysis and decision-making.

Reconstructs email conversation threads by linking related messages (via In-Reply-To, References headers, and subject matching), then aggregates thread context into a single coherent narrative. Implements thread reconstruction logic that handles provider-specific threading models (Gmail's conversation model vs. traditional IMAP threading) and presents full context to LLMs for holistic conversation understanding.

Implements rule-based email filtering using criteria (sender, subject patterns, content keywords, labels) to categorize and organize emails automatically. Rules are defined declaratively and executed server-side, applying labels, moving messages to folders, or marking as read based on matching conditions. Integrates with LLM decision-making to suggest or execute rules based on conversation context.

Assigns priority or importance scores to emails using heuristics (sender reputation, subject keywords, recipient list size, response time expectations) and optional ML models. Scores are computed server-side and exposed via MCP, allowing LLMs to reason about email importance for triage, response prioritization, and inbox management decisions. Integrates with user feedback to refine scoring over time.

Generates email draft suggestions based on conversation context, recipient information, and user preferences. Uses LLM capabilities (via Claude or other models) to compose natural-language email responses, subject lines, and full messages. Integrates with email context retrieval to ensure drafts reference previous conversation history and maintain tone/style consistency.